The case of a globally aphasic patient (JCU) is reported. Four experiments are described. In the first, she was able to make correct non-verbal semantic judgements about pictures. In naming, correct phonemic cues elicited correct names for 49% of the pictures. Incorrect cues elicited a variety of errors; semantic errors were the most frequent, and they were not often rejected by JCU saying “no” immediately afterwards. Unrelated word paraphasias were usually rejected. Cued word reading was no worse than naming but there were many fewer semantic errors. In a comprehension task, JCU was able to judge that correct names were correct, and that unrelated names were inappropriate, but she judged more than 50% of semantic co-ordinates of the correct names to be correct. The nature of the processes responsible for her semantic errors is discussed, and it is argued that JCU is using incomplete semantic information in name retrieval and word comprehension, but that she does not have a deficit specific to any particular lexical items.