In three parallel experiments, subjects rated their impressions of hypothetical 20-year-old males described by means of single or combined personality trait adjectives on 36 bipolar (Semantic Differential) scales. These ratings were a) intercorrelated across the scales (Q-technique) and factor analyzed separately for each subject, and b) submitted to an analysis of variance separately for each scale. It was shown that 1) when contradictory adjectives were combined the major portion of the variance of judgments fell on dimensions independent of that which characterized the contradiction; 2) there were substantial inter-individual consistencies in the direction of the deviation from that dimension; and 3) the utility of a linear model for the prediction of judgments of combined adjectives increases with increases in the extent to which the subject experiences the combination as consistent. However, the utility of the linear model decreases considerably and consistently if the combined adjectives are experienced as non-consistent.