Simple discriminations are involved in many functional skills; additionally, they are components of conditional discriminations (identity and arbitrary matching-to-sample), which are involved in a wide array of other important performances. Many individuals with severe disabilities have difficulty acquiring simple discriminations with standard training procedures, such as differential reinforcement. Errorless training methods may be more effective with this population. We used multiple-probe designs to compare two potentially errorless procedures for teaching simple discriminations among three pairs of photos of preferred items (S+) and colored rectangles (S-) to three youths with severe disabilities. In Experiment 1, baseline trials conducted with differential reinforcement yielded near-chance performances on all stimulus sets. A progressive delayed prompt training procedure was then implemented, with stimuli presented flat on the tabletop for one participant and at a 45-degree angle to the tabletop for the other participants. After 120 teaching trials, accuracy remained near chance. Next, a stimulus control shaping procedure was implemented using an adapted Wisconsin General Test Apparatus (WGTA), with stimuli at a 45-degree angle to the tabletop. Accuracy increased when this procedure was implemented with each stimulus pair in succession. In Experiment 2, for the participant whose stimuli were presented flat on the tabletop during the progressive delayed prompt training procedure, baseline trials were presented on the WGTA as at the end of Experiment 1, with differential reinforcement; accuracy remained high. On probe trials with stimuli placed flat on the tabletop, accuracy decreased to near-chance levels, indicating that the orientation of the stimulus array was a controlling variable.