Cognitive impairment is prevalent in at-risk mental states (ARMS) for psychosis.We studied cognitive functioning at baseline in ARMS individuals and investigated its power to predict ARMS persistence and remission at 2-year follow-up.196 patients were recruited. At baseline the ARMS population included 26 subjects meeting basic symptom (BS) criteria and 73 subjects fulfilling ultra-high risk (UHR) criteria. Two control groups were defined: 48 patients in a first episode of psychosis (FE), and 49 help-seeking patient controls (PCO). In 144 patients follow-up data were obtained. The 2-year risk of conversion to psychosis was 20%. Remission from an initial UHR state occurred in two thirds of the follow-up sample. UHR patients that converted to psychosis or did not remit during the follow-up (UHR(n-rem)) showed similar impairment in global cognitive functioning at baseline as the FE group, whereas global cognitive functioning in UHR patients with subsequent remission (UHR(rem)) approximated performances of the BS and PCO groups. UHR(n-rem) and UHR(rem) patients differed significantly on immediate verbal memory, but showed similarly impaired executive functions. Normal immediate verbal memory uniquely predicted remission from an at-risk state with a positive predictive value of 82%.Cognitive deficits are a characteristic feature of true ARMS patients. Verbal memory function appears critical in determining outcome.