We found evidence of bias due to matrix effect in 70% of 37 instrument/reagent-specific systems analyzing the total cholesterol content of a lyophilized proficiency testing material. We used a computational method to remove bias due to matrix effect from the proficiency testing database. After correction for matrix effect bias and when compared with the reference method, 92% to 93% of results for three lyophilized proficiency testing samples analyzed in 1989 and 1990 met the 1992 National Cholesterol Education Program total error goal of 8.9%, and 94% to 95% met the Clinical Laboratory Improvement Amendments of 1988 (CLIA '88) goal of 10%. However, compared with the definitive method for total cholesterol, the calibration bias of 41% of 37 peer groups exceeded the 1992 National Cholesterol Education Program goal for bias of 3%. Because the calibration bias of the method is incorporated into the peer group mean, use of peer group means as target values to assess result acceptability hinders advancement of the state of the art in interlaboratory comparability and the clinical effectiveness of laboratory testing. The prevalence of matrix effects has prevented successful application of accuracy-based evaluation of cholesterol test proficiency. The establishment of predictable recovery, preferably complete recovery, of cholesterol from reference materials is an important priority for cholesterol test methods. However, adjustment of proficiency testing results to remove the average bias due to matrix effects can help assess the actual state of the art in cholesterol test accuracy.