Background: Although patients with schizophrenia display an overall probabilistic category learning performance deficit, the extent to which this deficit occurs in unaffected siblings of patients with schizophrenia is unknown. There are also discrepant findings regarding probabilistic category learning acquisition rate and performance in patients with schizophrenia. Methods: A probabilistic category learning test was administered to 108 patients with schizophrenia, 82 unaffected siblings, and 121 healthy participants. Results: Patients with schizophrenia displayed significant differences from their unaffected siblings and healthy participants with respect to probabilistic category learning acquisition rates. Although siblings on the whole failed to differ from healthy participants on strategy and quantitative indexes of overall performance and learning acquisition, application of a revised learning criterion enabling classification into good and poor learners on the basis of individual learning curves revealed significant differences between percentages of sibling and healthy poor learners: healthy (13.2%), siblings (34.1%), patients (48.1%), yielding a moderate relative risk. Conclusions: These results clarify previous discrepant findings pertaining to probabilistic category learning acquisition rate in schizophrenia and provide the first evidence for the relative risk of probabilistic category learning abnormalities in unaffected siblings of patients with schizophrenia, supporting genetic underpinnings of probabilistic category learning deficits in schizophrenia. These findings also raise questions regarding the contribution of antipsychotic medication to the probabilistic category learning deficit in schizophrenia. The distinction between good and poor learning might be used to inform genetic studies designed to detect schizophrenia risk alleles. © 2010 Society of Biological Psychiatry.