One hundred twenty-eight observers randomly assigned to 16 treatment conditions in a modified Latin square design, viewed three videotapes of simulated classrooms in which teacher behavior was controlled (paralleling psychophysical procedures) to fit unambiguously into specific categories on ratings of frequency and variety of social reinforcement. True behavior deimension scores, person and performance consistency, and inferential level of coding forms were manipulated to determine their effects on frequency and variety ratings and six semantic differential items. Results of multivariate analyses of variance indicated that stimulus variables and observation system characteristics significanlty affected the mean, variance, and accuracy of observers' judgments. (Author)