A series of six experiments examined delayed identity matching‐to‐sample performances of subjects with mental retardation. The stimuli were either one or two simultaneously displayed forms. When the reinforcement contingencies required that only one form exert discriminative control, all subjects achieved high accuracy scores. However, accuracy scores were substantially lower when the contingencies required discriminative control by two forms, suggesting restricted stimulus control. The decline in matching accuracy appeared to reflect selective losses of conditional control by sample stimuli and shifts in control to features of the comparison stimulus displays. The experiments suggest improved techniques for assessing control by complex stimuli and for evaluating the effects of procedures that seek to broaden restricted stimulus control. The results challenge interpretations based on stimulus‐generalization decrement or shared atte