A theory is discussed in which observed performance ratings are derived from the distance between a rater reference point and subject performance point located on a postulated equal-interval scale and a postulated s-shaped rater characteristic curve, operationalized as the normal ogive. Least-squares estimates of rater (nR=47, 31, and 29) and subject (nS=29, 30, and 35) points were determined separately on each of three junior medical student cohorts' data. The proposed model fit each of the data sets better than two alternative models (r is greater than 0.70; p is less than 0.01). Test-retest reliabilities for rater parameters were r is less than 0.29 (joint p is less than 0.04). Cross-validating results also supported the theory. (Author/RL)