Severely retarded adults (51) were taught a difficult visual discrimination in an assembly task by 1 of 3 training techniques: adding and reducing large cue differences on the relevant-shape dimension; adding and fading a redundant-color dimension or a combination of the 2 techniques. There were significant differences between training conditions in both trials- and errors-to-criterion performance. These were attributed to differential establishment of stimulus control in the 1st phase of training, with the combined procedure being the most effective, the color coding fading next and the relevant-dimension cue-disparity method the least powerful. Substantial differences may exist in the efficacy of various vocational-skill training procedures involving manipulation of stimulus features.