The antisaccade task is a promising schizophrenia endophenotype; it is stable over time and reflects neurophysiological deficits present in both schizophrenia subjects and their first-degree relatives. Meaningful genetic research requires large sample sizes that are best ascertained using multi-site study designs. To establish the criterion validity of the antisaccade task in a multi-site design, the Consortium on the Genetics of Schizophrenia (COGS) examined whether seven sites could detect previously reported antisaccade deficits in schizophrenia subjects. Investigators presented 3 blocks of 20 antisaccade stimuli to 143 schizophrenia subjects and 195 comparison subjects. Frequent collaborator communication, standardized training, and ongoing quality assurance optimized testing uniformity. Data were discarded from only 1.2% of subjects due to poor quality, reflecting the high fidelity of data collection and scoring methods. All sites detected a significant difference in the proportion of correct antisaccades between schizophrenia and comparison subjects (p < .02 at all sites); group differences in gain and latency were less robust. Regression analyses to adjust for the effects of group, site, age, gender, smoking, and parental education on the proportion of correct antisaccades revealed a significant effect of group, site, and age but no effect of gender, smoking, or parental education, and no group-by-site interactions. Intraclass correlations between proportion of correct antisaccades across the blocks of stimuli ranged from 0.87 to 0.93, demonstrating good within-session reliability at sites. These results confirm previous findings of antisaccade deficits in schizophrenia subjects and support the use of the antisaccade task as a potential schizophrenia endophenotype in multi-site genetic studies.