Generally, ratings have notoriously low inter-rater reliabilities. Because of differences in orientation, background, and expectations, ratings are seldom made from the same point of reference; thus, many types of error mask the true rating variance. Guilford's technique identified most types of constant error by analysis of variance and then eliminated by an adjustment technique. This, however, does not remove all possible errors; it only "cleans up" the ratings to the extent these known errors no longer contribute to the error variance. The technique, moreover, is not applicable where not all raters have ratings for all ratees on all traits. It is proposed that if the number of missing values is relatively small, one of several methods be used to estimate the missing data before an analysis of variance is performed. Depending on the validity of the statistical assumptions made in the estimation, these methods are capable of producing reasonable estimates for the missing data. Three possible methods are illustrated with data taken from Guilford's work. The accuracy of the methods is compared before performing the analysis of variance and making the required adjustments. Further, the inter-rater reliabilities of the adjusted data and of the unadjusted data are estimated, using the Spearman-Brown formula by analysis of variance. A comparison of the two reliability coefficients reflects the degree to which the adjustment has been useful. (Author)