This study tested the relative predictive powers of three theories of schizophrenia: Mednick's drive theory; Rodnick and Garmezy's censure-deficit theory; and Atkinson and Robinson's censure-sensitivity theory. Thirty normals, 30 process schizophrenics, and 30 reactive schizophrenics were each run in a low-and high-complexity verbal discrimination task. Half of each group was praised for correct responses, the other half was censured for incorrect ones. While normals did not respond differentially to the reinforcers, schizophrenics tended to learn faster when censured than when praised, supporting Atkinson and Robinson. The performance decrement from the low-to the highcomplexity task was equivalent for all groups, which is nonsupportive of Mednick's theory. An alternative explanation of the results is offered and a cautious interpretation of the data is urged due to medication differences between normals and schizophrenics.