Word-finding difficulties of dysphasic patients are assisted by providing semantic or phonological cues about objects and their names. Experiments that have evaluated a single application of such cues suggest that semantic assistance is more beneficial. However, both approaches seem helpful in repeated therapy sessions. An experiment to identify the source of this assistance was performed in 7 patients with dysphasic difficulties following strokes. Therapy consisted of a minimum of 10 sessions spread over approximately 5 wks. Results replicated reports that therapy effects persist for long periods but failed to find a difference in the effects of word/pictures-matching tasks where semantic information must be accessed and a task where the picture was seen while its name was repeated. An unexpected finding was that generalization to nontarget items in the therapy task occurred, but only for those unrelated to the targets.