Saccadic eye movements are well-described markers of cerebral function and have been widely studied in schizophrenia spectrum populations. However, less is known about saccades in individuals clinically at risk for schizophrenia. Therefore, we studied individuals in an at-risk mental state (ARMS) (N = 160), patients in their first episode of schizophrenia (N = 32) and healthy controls (N = 75). N = 88 ARMS participants showed an early at-risk mental state (E-ARMS), defined by cognitive-perceptive basic symptoms (COPER) or a combination of risk and loss of function, whereas N = 72 were in a late at-risk mental state (L-ARMS), defined by attenuated psychotic symptoms or brief limited intermittent psychotic symptoms. We examined prosaccades, reflecting overt attentional shifts, and antisaccades, measuring inhibitory control, as well as their relationship as an indicator of the interplay of bottom-up and top-down influences. L-ARMS but not E-ARMS participants had increased antisaccade latencies compared to controls. First-episode patients had higher antisaccade error rates compared to E-ARMS participants and controls, and increased latencies compared to all other groups. Prosaccade latencies did not differ between groups. We observed the expected negative correlation between prosaccade latency and antisaccade error rate, indicating that individuals with shorter prosaccade latencies made more antisaccade errors. The magnitude of the association did not differ between groups. No saccadic measure predicted conversion to psychosis within 2 years. These findings confirm the existence of antisaccade impairments in patients with schizophrenia and provide evidence that volitional response generation in the antisaccade task may be affected even before onset of clinically overt psychosis.