Poor performance on the antisaccade task has been proposed as a candidate endophenotype in schizophrenia. Caveats to this proposal, however, include inconsistent findings in first-degree relatives of individuals with schizophrenia, and substantial heterogeneity in individuals with the disorder. In this study, we examined antisaccade performance in patients and relatives, and sought to establish whether antisaccade measures could differentiate between two patients clusters identified in the Western Australian Family Study of Schizophrenia with either pervasive cognitive deficits (CD) or cognitively spared (CS). Ninety-three patients (CD = 47, CS = 46), 99 relatives and 62 healthy controls carried out a standard antisaccade task. Results showed: (i) significantly greater error rate, and prolonged latencies to correct responses and self-correction saccades in patients compared with controls; (ii) high error rates in relatives of poorly performing patients; (iii) longer latencies of self-correction saccades in relatives compared to controls; and (iv) higher error rate and longer latencies of self-correction saccades in the CD subgroup compared with CS. Unaffected relatives as a group were unimpaired in error rate as compared to healthy controls. These findings suggest that the antisaccade error rate and latency of self-correction saccades are useful measures in specific applications of genetic research in schizophrenia, without fully meeting endophenotype co-familiality requirements. (Contains 4 tables and 3 figures.)