Two nonfluent aphasic patients, ML and AR, and two fluent aphasic patients, LC and LW, along with older controls were tested on tasks involving semantic blocking and lexical selection. In Experiment 1, a repeated picture-naming task manipulating semantic blocking and presentation rate, both nonfluent patients showed growing semantic interference across trials, while the other subjects did not. In Experiment 2, a corresponding comprehension task, ML showed increasing semantic interference while AR showed a lack of repetition priming. Experiment 3 was a category fluency task in which AR performed very poorly and ML performed similarly to the fluent patients. Experiment 4 was a sentence completion task, and no evidence was found that the number or relatedness of lexical competitors caused any difficulty for the nonfluent patients. It was concluded that ML suffers from deficient semantic inhibitory mechanism while AR suffers from an under-activation of lexical representations.