Evaluated were variables effecting the differential performance of 627 educable mentally retarded Ss (mean age 14.5 years) on a test-train-retest task designed to measure learning potential. Family, social, health, schooling, and testing data were collected for the Ss of which 75% were students in public school special classes and 25% were residents of state institutions for the retarded. The test designs of the Kohs Block Designs were administered individually three times: prior to coaching, 1 day following, and 1 month following coaching. Findings indicated that the learning potential procedure resulted in a stable assessment of ability to reason (as seen in the high correlation between immediate posttest and delayed posttest), that 23% of the variance in the delayed posttest was uniquely predicted by training, that the training procedure did not elicit systematic differences in scores from children of different racial and social class backgrounds, that the nonverbal block designs permitted some students from poor language backgrounds to demonstrate ability to reason, and that greater than expected effects of training were demonstrated by children from large families, Southern-born blacks, and foreign-born children. (DB)