concurring in part and dissenting in part:

I agree with Part II of the majority opinion, explicating the policymaking authority of the Department of Revenue, determining that the “20% rule” cannot be used in administrative driver’s license revocation proceedings absent compliance with the rulemaking procedures of section 24-4-103, 10A C.R.S. (1988), and concluding that the hearing officer erred by applying the “20% rule” to resolve the conflicting results of the two intoxilyzer tests involved in this case. However, I respectfully dissent from Part III of the majority opinion. In my view, the hearing officer acted in an arbitrary and capricious manner by summarily applying the “20% rule” and by failing to evaluate by appropriate standards the discrepancy between the two test results. Accordingly, I would remand to the court of appeals with directions to return the case to the Department of Revenue for a determination of whether, applying the proper standards, Robinson’s blood-alcohol content was 0.15 or higher.

In administrative revocation proceedings, judicial review of hearing officer decisions is limited to an evaluation of whether the Department of Revenue “exceeded its constitutional or statutory authority, made an erroneous interpretation of the law, acted in an arbitrary and capricious manner, or made a determination which is unsupported by the evidence in the record.” § 42-2-122.1(9)(b), 17 C.R.S. (1984). “Where conflicting evidence is presented in an administrative hearing, the credibility of witnesses and the weight to be given their testimony are decisions within the province of the agency.” Charnes v. Lobato, 743 P.2d 27, 32 (Colo. 1987); see also G & G Trucking Co. v. Public Util. Comm’n, 745 P.2d 211, 216 (Colo. 1987); Acme Delivery Serv. v. Cargo Freight Sys., 704 P.2d 839, 843 (Colo. 1985). When acting in its capacity as the trier of fact, however, an administrative agency is entitled to deference only when it applies the correct criteria in resolving evidentiary conflicts. See Gonzales v. Industrial Comm’n, 740 P.2d 999, 1001 (Colo. 1987). Therefore, inquiry into whether the agency applied the proper principles of law in evaluating the evidence before it is not forestalled simply because the record supports the agency’s ultimate findings. Id.; see also Electric Power Research Inst. v. City and County of Denver, 737 P.2d 822, 825-26 (Colo. 1987).

In Lobato, we upheld a hearing officer’s exercise of discretion in determining that although the conflicting results of intoxi-lyzer tests weighed evenly, the preponderance of the evidence issue should be resolved in favor of the test given nearer to the time that the offense of driving under the influence of alcohol was committed. Lobato, 743 P.2d at 30-34. In reaching this conclusion we noted that

*70[m]any factors may enter into a hearing officer’s decision that one test is more reliable than another, including but not limited to such variables as the equipment and chemical standards used, the experience of each witness, and the circumstances surrounding a particular test. The determination as to which test result has greater reliability and deserves greater weight is within the agency’s discretion and not an issue to be analyzed on appellate review.

Lobato, 743 P.2d at 33.

In the instant case, the majority opinion concludes that despite the hearing officer’s erroneous application of the “20% rule,” there was sufficient evidence in the record upon which the hearing officer could make a determination that Robinson drove a vehicle with at least a 0.15 blood-alcohol content. In my opinion, the record contains no indication that the hearing officer based his determination on anything other than the results of the two intoxilyzer tests, and the “20% rule.” Indeed, the transcript of the hearing officer’s ruling on the blood-alcohol content issue consists of only a few remarks directed toward the discrepancy between the two intoxilyzer tests. These remarks reveal that the hearing officer found that the results of the second intoxilyzer test supported rather than refuted those of the first test. The hearing officer clearly reached this conclusion by operation of the “20% rule.” See maj. op. at p. 64 (quoting full text of hearing officer’s remarks on the blood-alcohol content issue).

We have no way of knowing on the basis of this record whether the hearing officer would have reached the same ultimate result without the benefit of the “20% rule.” Cf. Salt River Project v. United States, 762 F.2d 1053, 1060-61 n. 8 (D.C.Cir. 1985) (affirming agency action without remanding where agency clearly would have reached the same ultimate result had it relied on only those findings that were not erroneous). Nothing in the record suggests that the hearing officer gave any consideration to the types of factors deemed appropriate in Lobato for resolving conflicts between intoxilyzer test results, or to any other evidence concerning blood-alcohol content.1 Consequently, it is a usurpation of the agency’s role for this court to assess credibility and weigh testimony or other evidence in reaching a determination that the record supports a finding by a preponderance of the evidence that Robinson’s blood-alcohol content was 0.15 or higher. I would therefore hold that the hearing officer’s determination to revoke Robinson’s license was arbitrary and capricious and remand to the court of appeals with directions to return the case to the Department of Revenue for reconsideration.

ERICKSON, J., joins in this concurrence and dissent.

The Motor Vehicle Division of the Colorado Department of Revenue (DMV) appeals from the court of appeals’ unpublished decision in Robinson v. Charnes, No. 85CA1542 (Colo.Ct.App. June 4, 1987) which reversed the district court’s affirmance of DMV’s revocation of Larry L. Robinson’s driver’s license. We accepted cer-tiorari to determine whether the hearing officer properly relied on the so-called “20% rule” 1 to support his finding that Robinson had driven a motor vehicle with a blood alcohol content of .150 or higher. We agree with the court of appeals that the “20% rule” cannot be relied upon in revocation hearings. However, because we find that there was sufficient evidence in the record exclusive of the “20% rule” to support the hearing officer’s determination to revoke Robinson’s driving license, we reverse the judgment of the court of appeals and direct that the judgment of the trial court be reinstated.

I.

On November 7, 1984, Robinson was stopped by Officer Clifford Thompson of the Columbine Valley Police Department for failing to stop at a stop sign. On approaching Robinson and requesting his driver’s license and automobile registration, Officer Thompson detected a strong *64odor of an alcoholic beverage on Robinson’s breath. When a backup officer summoned by Thompson arrived, Robinson was directed to perform roadside maneuvers which he was unable to complete satisfactorily.

Robinson was arrested for driving under the influence and transported to the Arapahoe County Sheriffs Department. Robinson submitted to an intoxilyzer breath test which was conducted at 10:24 p.m., within an hour of the traffic stop. The test result indicated Robinson’s blood alcohol content (BAC) to be .181 grams of alcohol per 210 liters of breath. A second breath sample was taken and preserved for Robinson. Robinson then was issued a summons for driving under the influence and given a notice of revocation of his driver’s license.

Robinson contested the revocation at a hearing held on January 7, 1985. At the hearing, Officer Thompson testified to the facts which occurred at the time of the arrest, including the .181 BAC result of the intoxilyzer test. Officer Thompson also identified a packet of documents labeled as Exhibit A which included the notice of revocation, the summons and complaint issued to Robinson, an offense report providing a brief summary as to what happened at the traffic stop, an evidence receipt for the silicone gel tube containing Robinson’s preserved breath sample, the intoxilyzer printout showing the result of .181 BAC, an Alcohol Influence Report which briefly summarized the officer’s observation of the roadside maneuvers, and a checklist completed by the intoxilyzer operator while administering the intoxilyzer test.

Dr. Lawrence Kier, recognized as an expert in the area of toxicology of blood and breath analysis, testified on behalf of Robinson that he had analyzed the second breath sample according to Colorado Department of Health regulations and obtained test results of .149 grams of alcohol per 210 liters of breath. In addition to his test result findings, Kier also testified about potential sources of error which might explain the discrepancy between the test results obtained at the time of the arrest and the test result which he had obtained from the sample preserved for Robinson.

After the testimony by Kier, Robinson’s only witness, and the closing argument of Robinson’s counsel, the hearing officer addressed the issues raised by counsel. With respect to the discrepancy between the two intoxilyzer test results, the hearing officer stated:

As to the matter of the second test result obtained, this office has been personally informed by a professional colleague of Dr. Kier’s, a gentleman also equally recognized as an expert in the same field in this state, Dr. Dale C. Wingeleth, that a second test result within 20% of the first test result supports that first test result. It does not refute it. When the matter came up, I did a little quick arithmetic, gentlemen. A result 20% below .181 would be a .1448. The result obtained here, .149 is within that range, and as I consider it, as supporting the result obtained, not refuting it.

The hearing officer discussed the other issues raised and concluded: “Having therefore heard the testimony and evidence that has come forward at this hearing and the arguments raised by Counsel regarding the matter ... I would in preponderance find that the evidence would sustain revocation of the respondent’s license and driving privilege....”

Robinson sought judicial review in the district court which affirmed the hearing officer’s decision. The court of appeals reversed the revocation order and remanded the case with instructions to reinstate Robinson’s driver’s license.

The court of appeals ruled that the use of the “20% rule” to bolster the reliability of the first test was impermissible and, for that reason, the hearing officer’s ruling was arbitrary and capricious. The court of appeals further ruled that, because the test results were of equal weight, the DMV failed to meet its burden of proof and therefore the revocation could not stand.

We granted certiorari to review this case because there appear to be inconsistent rulings by the court of appeals concerning application of the “20% rule.” In Schocke *65v. Department of Revenue, 719 P.2d 361 (Colo.Ct.App. 1986), the hearing officer applied the “20% rule.” Finding that the variance between the two test results was less than 20%, the hearing officer refused to consider the results of the second test and revoked the driver’s license. Id. at 362-363. The court of appeals reversed. It found that both tests were presumed correct because both had been conducted in accordance with the Colorado State Department of Health rules and regulations. Id. at 363. Because no evidence other than the test results was introduced and because the evidence weighed evenly, the court of appeals held that the department failed to meet its burden of proof by the preponderance of the evidence. Therefore, the court ruled that the hearing officer erred in revoking the license.2 Id. In Harvey v. Charnes, 728 P.2d 373 (Colo.Ct.App. 1986), the court of appeals ruled that it was improper to measure the reliability of the intoxilyzer test performed by a police officer by a “20% rule” because the rule was based on evidence outside the record. However, the error was held to be harmless since there was sufficient evidence of excessive blood alcohol level during the requisite period, i.e., one BAC test exceeded .15, the driver was driving in an erratic manner, she had the odor of an alcoholic beverage on her breath, she slurred her speech, and she failed the roadside sobriety test. Id. at 374. In the present case, the court of appeals found application of the “20% rule” to be arbitrary and capricious but did not apply a harmless error analysis as it did in Harvey. We will consider both the validity of the application of the “20% rule” in driver’s license revocation hearings and the application of harmless error in this context.

II.

The origins of the “20% rule” are uncertain at best. In Schocke, 719 P.2d at 362, the hearing officer referred to the “20% rule” as a department policy. Here, the hearing officer indicated only that he was personally informed by an “expert” as to the rule. We have been directed to no professional literature supporting this rule and how this methodology came into existence for resolving differences between in-toxilyzer test results is not readily apparent. Other than the brief mention in Schocke, Harvey, and the present case, we have found no reference to the “20% rule” or similar statistical methodologies that direct how different test results will be interpreted when considered as evidence in a revocation hearing.

It is clear, however, that the “20% rule” has been applied by hearing officers as having the force and effect of law. In order to be so applied, the “20% rule” must be authorized by a statute, rule or decision. The governing statutes, however, contain no such principle. Administrative proceedings involving the revocation of drivers’ licenses are subject to the State Administrative Procedure Act (APA). See § 42-2-122.1(10), 17 C.R.S. (1984). Under section 24-4-104(2), 10A C.R.S. (1988) of the APA, any agency making a determination as to the revocation of a license shall do so “based solely upon stated criteria, terms, and purposes of the statute, or regulations promulgated thereunder, and case law interpreting such statutes and regulations pursuant to which the license is issued or required.”

Section 42-2-122.1(8)(c), 17 C.R.S. (1984) which provides the statutory criteria for mandatory revocation of a license pursuant to section 42-2-122.1(l)(a)(I), 17 C.R.S. (1984) states only that:

The sole issue at the hearing shall be whether by a preponderance of the evidence the person drove a vehicle in this state when the amount of alcohol in such person’s blood was 0.15 or more grams of alcohol per 100 milliliters of blood or 0.15 or more grams of alcohol per 210 liters of breath at the time of the commission of the alleged offense as shown *66by chemical analysis of such person’s blood or breath_3

Since no “20% rule” or similar statistical procedure has been adopted by statute, the rule can be valid only if properly adopted through the agency’s rulemaking or adjudicative powers.

No regulation embodying the “20% rule” has been promulgated by the Director of the Department of Revenue pursuant to his rulemaking authority in section 42-1-204, 17 C.R.S. (1984). The “20% rule” thus appears to have been developed and applied solely in the course of revocation hearings.

We recognize that, in general, an agency may make policy through either its adjudicatory or its rulemaking power. However, in Home Builders Association v. Public Utilities Commission, 720 P.2d 552 (Colo. 1986), we noted that the agency’s discretion to choose between rulemaking and adjudication is limited. We held that the Public Utility Commission’s adoption of a method for calculating the amount of a monetary deposit in an adjudicatory proceeding was nothing less than an “agency statement of general applicability and future effect implementing [and] declaring policy.” As such, the policy was a rule or regulation within the meaning of the APA and the agency decision was void for lack of compliance with the rulemaking procedures of the APA. Id. at 561. See also Crema v. New Jersey Dep’t of Envtl. Protection, 94 N.J. 286, 463 A.2d 910 (1983) (where contemplated action is intended to have wide application and prospective effect, rulemaking becomes the suitable mode of proceeding).

In Mayberry v. University of Colo. Health Sciences, 737 P.2d 427, 430 (Colo.Ct.App. 1987), the court of appeals considered policymaking by adjudication and invalidated a change in Personnel Board policy because the change was made by the board in an adjudicatory hearing without resort to formal rulemaking. At issue was the board’s interpretation of one of its rules. Where the board had given the rule a prior consistent interpretation which had been upheld on appeal, the court of appeals held that the board could not change that interpretation in the course of adjudicating an individual employee’s appeal of her firing. The board was required to proceed by rulemaking if it wished to change its rule. Other jurisdictions have reached the same result. See, e.g., Department of Labor v. Titan Constr. Co., 102 N.J. 1, 504 A.2d 7, 17 (1985) (invalidating adjudicatory expansion of department of labor department provisions to corporate officers); Crema, 463 A.2d 910 (invalidating adjudicatory adoption of a conceptual review and approval process for proposed residential developments). Underlying these decisions is a concern for the fairness of the adjudicatory process. Fairness requires that rule-making power be used to establish standards which an agency intends to enforce by adjudication so that all parties to the proceeding know the “rules of the game.” Titan Constr. Co., 504 A.2d at 14.

Policymaking in the adjudicatory setting serves principally to provide a guide to the agency’s position in future adjudicatory proceedings. See NLRB v. Bell Aerospace Co., 416 U.S. 267, 294, 94 S.Ct. 1757, 1771-72, 40 L.Ed.2d 134 (1974); NLRB v. Wyman-Gordon Co., 394 U.S. 759, 764-766, 89 S.Ct. 1426, 1428-1430, 22 L.Ed.2d 709 (1968) (plurality opinion). Establishment of policymaking through adjudication is justified in circumstances where an agency must treat matters neither anticipated nor dealt with previously by the agency or matters that are extremely complex and incapable of being reduced to a formalized statement of policy. See SEC v. Chenery Corp., 332 U.S. 194, 202-203, 67 S.Ct. 1575, 1580-1581, 91 L.Ed. 1995 (1947). Policy-making is done through adjudication when “[i]t is doubtful whether any generalized standard could be framed which would have more than marginal utility.” Bell Aerospace Co., 416 U.S. at 294, 94 S.Ct. at 1772.

*67This description of complex, unanticipated fact patterns which cannot be encompassed in a general rule does not fairly characterize DMV revocation hearings. When agency decisions such as DMV revocation hearings are not published and are not readily available to the public, such adjudicatory proceedings are not the appropriate forum for development of agency policy. Parties to such administrative hearings are not likely to be aware of prior cases in which agency criteria and standards were adopted and, as a result, will be unable to adequately prepare or properly present their cases. One commentator has stated:

[Without knowledge of agency criteria and standards, lawyers representing licensing clients cannot perform their functions properly. If decision factors are unknown, a court appeal may be virtually useless in states where judges sustain administrative decisions on the basis of substantial evidence in the administrative record reviewed. But the attorney is assisted in protecting his client’s interests if criteria are known, for he may obtain judicial evaluation of the criteria or their application by the agency to the facts of his case.

J. Reese, Power, Policy, People: A Study of Driver Licensing Administration 186 (1971) (footnote omitted).

The adoption of the “20% rule” was not appropriately done in an adjudicatory setting because it was foreseeable by the DMV that conflicting evidence of breathalyzer results would be introduced in revocation hearings and that the triers of fact would be required to determine how to resolve the differences. Here, the statutory scheme specifically provides that, in the prosecution for driving under the influence of alcohol, “the defendant shall be entitled to offer direct and circumstantial evidence to show that there is a disparity between what the tests show and other facts so that the trier of fact could infer that the tests were in some way defective or inaccurate.” § 42-4-1202(1.5)(b), 17 C.R.S. (1988 Supp.). It also was foreseeable that the agency’s adoption of a methodology to resolve test differences would result in a rule of general applicability to future cases involving the same issue of conflicting test results. Foreseeability of future application of an adopted methodology was a key finding in Home Builders Association which led us to hold that the policy adopted by the Public Utilities Commission in an adjudicatory proceeding was void for failure to comply with the APA rulemaking procedures set out in section 24-4-103. 720 P.2d at 562.

Whether a policy is established by either rulemaking or adjudication, each method requires some form of participation and input in determining the policy to be adopted. Section 24-4-103, 10A C.R.S. (1988), of the APA requires that when a regulation is promulgated through rule-making, a public announcement is made of the proposed rulemaking, a public hearing is held, and interested persons are afforded an opportunity to submit written data, views and arguments. Only after this process is followed and the information obtained is the agency permitted to adopt the proposed regulation.

With regard to participation and input in adjudication, section 24-4-105, 10A C.R.S. (1988) provides in pertinent part that:

(7) ... [Ejvery party to the proceeding shall have the right to present his case or defense by oral and documentary evidence, to submit rebuttal evidence, and to conduct such cross-examination as may be required for a full and true disclosure of the facts.

[[Image here]]

(8) An agency may take notice of general, technical, or scientific facts within its knowledge, but only if the fact so noticed is specified in the record or is brought to the attention of the parties before final decision and every party is afforded an opportunity to controvert the facts so noticed.

Because the “20% rule” was not promulgated pursuant to statutory rulemaking procedures, there was no opportunity for the public to challenge the basis and purpose of the rule. Similarly, in this adjudi*68catory proceeding, Robinson was not able to refute or attack the statistical premise or the validity of application of the “20% rule,” i.e., whether a variance of less than 20% leads to a conclusion that two different test results are supportive of one another or whether the specific statistical application is appropriate, given the type of data presented. The hearing officer, in effect, took judicial notice of an alleged scientific fact that “a second test result within 20% of the first test result supports that first result.” Contrary to section 24-4-105(8), Robinson was not afforded an opportunity to controvert the fact noticed.

In summary, the “20% rule” is a rule or regulation within the meaning of the APA and because it was not promulgated according to the rulemaking authority delegated to the Director of Revenue, it is invalid. The hearing officer erred by applying the “20% rule” to resolve differences between two intoxilyzer test results when he determined whether Robinson’s BAC was above the .15 level pursuant to section 42-2-122.1(l)(a).

III.

Even though the hearing officer improperly used the “20% rule” to support his findings, the DMV argues that there was substantial evidence within the record to support the findings of the hearing officer that Robinson had driven a motor vehicle with a blood alcohol content of .15 or higher. On this basis, the DMV urges us to uphold the revocation of Robinson’s license. We agree.

The statutory standard for judicial review provides that a reviewing court may reverse the DMV’s revocation of a license if it finds that the “department exceeded its constitutional or statutory authority, made an erroneous interpretation of the law, acted in an arbitrary and capricious manner, or made a determination which is unsupported by the evidence in the record....” § 42-2-122.l(9)(b). To determine that a hearing officer’s decision was arbitrary and capricious, a reviewing court must be convinced from the record as a whole that there was not substantial evidence to support the hearing officer’s decision. DeScala v. Motor Vehicle Div., 667 P.2d 1360 (Colo. 1983); Glasmann v. Department of Revenue, 719 P.2d 1096 (Colo.Ct.App. 1986).

The question presented by this case is whether there is substantial evidence in the record to support the revocation of Robinson’s license, despite the hearing officer’s erroneous use of the “20% rule.” Here, our review of the record convinces us that there is sufficient evidence to prove the one salient fact, i.e., that Robinson drove a vehicle with .15 BAC or greater. The police performed the breath test in accordance with the required procedures and Robinson’s BAC was measured as .181. Evidence also was presented by Officer Thompson which showed that Robinson was stopped for committing a traffic infraction, he had the odor of an alcoholic beverage on his breath, his speech was slurred and he was unable to perform satisfactorily roadside physical maneuvers. See Harvey v. Charnes, 728 P.2d at 374. Thus, there was evidentiary support for the revocation of Robinson’s driver’s license.

Because we find the evidence is clearly sufficient to uphold the agency action revoking Robinson’s license, it is not necessary for us to reverse and remand the case for reconsideration by the hearing officer. A reviewing court’s power, under these circumstances to affirm the agency action without remanding, is well established:

When an agency relies on a number of findings, one or more of which are erroneous, we must reverse and remand only when there is a significant chance that but for the errors the agency might have reached a different result. When it is clear that based on the valid findings the agency would have reached the same ultimate result, we do not improperly invade the administrative province by affirming.

Salt River Project v. United States, 762 F.2d 1053, 1061 n. 8 (D.C.Cir. 1985). Accord Consolidated Gas Supply v. Federal Energy Regulatory Commission, 606 F.2d 323, 329 (D.C.Cir. 1979), cert. denied, 444 U.S. 1073, 100 S.Ct. 1018, 62 L.Ed.2d 755 *69(1980) (“The appropriate standard is to remand for correction of error only when there is substantial doubt that the administrative agency would have reached the result it did absent reference to the material.”). See also 2 Koch, Administrative Law and Practice § 8.12, at 7 (1987 Supp.) (“Indeed, the court should not remand when, even though it rejects some of the finding, there are valid findings upon which the agency would have reached the same ultimate result.”). Cf. O’Leary v. Brown-Maxon, Inc., 340 U.S. 504, 508, 71 S.Ct. 470, 472, 95 L.Ed. 483 (1951), where, in an opinion by Justice Frankfurter, the Supreme Court declined to remand the case to the administrative agency and the Court itself examined the record to assess the sufficiency of the evidence because “we have a slim record and the relevant standard is not difficult to apply; and we think the litigation had better terminate now.” Thus, we do not usurp the DMV’s authority by affirming the license revocation on the grounds suggested by the agency.

Accordingly, we conclude that the hearing officer’s determination to revoke the license was not arbitrary and capricious and that the court of appeals erred in reversing the revocation order. We remand the case to the court of appeals with directions to remand the case to the district court for reinstatement of the revocation order.