Two adjective pairs in comparative form, more-less and taller-shorter, were tested with moderately mentally retarded (MR) and nonretarded children (NR) matched on verbal mental age (MA) and socioeconomic status (SES) in a "show" and an "act-out" experimental condition. The data indicated that all the main effects, the group, the experimental condition, and the unmarked-marked relational terms were significant and so were the interactions of the Group x Experimental Conditions, Group x Unmarked Versus Marked Relational Terms, and Group x Experimental Conditions and Unmarked Versus Marked Relational Terms. A logistic regression analysis (LRA) showed that the relational term shorter in the act-out condition best predicted classification membership for the subjects in either group, suggesting that the MR children were less able to self-regulate behavior. The error types, however, mainly observed in the act-out experimental condition were the same from a semantic point of view in either group. The effect of the unmarked versus marked distinction cannot be considered independent of the experimental conditions applied.