Verbal fluency deficits in schizophrenia are difficult to interpret because the tasks are multi-factorial and groups differ in total words generated. We manipulated retrieval and switching demands by requiring alternation between over-learned sequences in which retrieval is relatively automatic (OS) and semantic categories requiring increased retrieval effort (SC). Controlled processing was also manipulated by including switching and non-switching conditions, and formal thought disorder (FTD) was assessed with the communication disorders index (CDI). The OS/SC semantic fluency paradigm was administered during fMRI to 13 patients with schizophrenia and 14 matched controls. Images were acquired on a 3 Tesla Siemens scanner using compressed image acquisition to allow for cued overt word production. Subjects alternated between OS, SC, OS-switch, SC-switch, and baseline blocks. Images were pre-processed in SPM-2, and a two-stage random effects analysis tested within and between group contrasts. There were no group performance differences. fMRI analysis did not reveal any group differences during the OS non-switching condition. Both groups produced expected activation in bilateral prefrontal and inferior parietal regions. However, during the SC condition patients had greater activation than controls in left prefrontal, right anterior cingulate, right superior temporal, bilateral thalamus, and left parietal regions. There was also evidence of patient over-activation in prefrontal, superior temporal, superior parietal, and visual association areas when a switching component was added. FTD was negatively correlated with BOLD response in the right anterior cingulate, cuneus and superior frontal gyrus during increased retrieval demand, and positively correlated with fMRI activation in the left lingual gyrus, right fusiform gyrus and left superior parietal lobule during increased switching demand. These results indicate that patients are able to successfully perform effortful semantic fluency tasks during non-speeded conditions. When retrieval is relatively automatic there does not appear to be an effect of schizophrenia on fMRI response. However, when retrieval and controlled processing demands increase, patients have greater activation than controls despite unimpaired task performance. This inefficient BOLD response may explain why patients are slower and less accurate on standard self-paced fluency tasks.