An instrument for the assessment of perception of nonverbal facial affect was developed and administered to two separate groups of respondents: adults with mental retardation and adults without mental retardation. The instrument was developed and calibrated using an item response theory (Rasch) analysis on respondents without mental retardation. Following assurance of item stability, data were analyzed using an anchored analysis for persons without mental retardation. Cumulative score differences between the two groups were expected and were found. The Rasch analysis uncovered a difference in the structure of the latent trait of understanding of facial affect between the two groups. In view of these qualitative differences, the argument is presented that quantitative differences in the two groups are irrelevant. We suggest that qualitative differences such as those found herein may partially account for the traditionally limited scope of generalization and maintenance of treatment effects of social skills training with persons with mental retardation. Theoretical and empirical implications of the findings, and future research directions based on these qualitative differences are discussed.