<i>Background:</i> Neurocognitive impairment has consistently been considered a central and stable feature in schizophrenia. As this possibility has been far less studied in bipolar disorder, we aimed to prospectively investigate the stability and specificity of cognitive performance in bipolar disorder compared to schizophrenia. <i>Methods:</i> Fifteen DSM-IV bipolar type I patients and 15 schizophrenic patients were assessed twice with a comprehensive neuropsychological battery and the Positive and Negative Syndrome Scale over a 3-year follow-up. The cognitive performance of the groups was compared at baseline and 3 years later as a mean with that of 26 healthy volunteers. Endpoint and baseline assessments were also compared for each patient group in order to evaluate the stability of cognitive impairment. <i>Results:</i> At both time points, bipolar and schizophrenic patients showed significant deficits on most of the cognitive tasks compared to healthy subjects. Overall, the cross-sectional cognitive profile was similar for both patient groups. Moreover, after controlling for age and length of illness, the two groups’ cognitive function did not differ over time in any test. With the exception of the Stroop color-word interference task, performance at baseline for each test but neither length of illness nor diagnostic category predicted the endpoint performance. <i>Conclusion:</i> This preliminary study suggests that cognitive impairment is also mainly stable over time in bipolar I disorder and thus not specific to schizophrenia.