A programmed training task based on the token test was designed and administered to 24 adult aphasics. The point at which the patient failed the token test determined where he began the program and how many training sessions he would receive, provided he did not exceed the error rate before he finished the program. Improvement and errors within the program were analyzed to evaluate this training. Mildly or mild to moderately impaired patients showed significant improvement, while moderately and severely handicapped aphasics did not. This training, however, did not generalize to other "direction-following" tasks for any group of subjects. Within the program, generalization items were the most difficult for all aphasics. Neither in- versus outpatient status nor frequency of training sessions were significantly related to success with this program. This training procedure appeared effective for mildly impaired aphasics. The only overlap between the two groups which emerged in this study involved the youngest subject whose aphasia was the result of a recent traumatic injury.