BACKGROUND: A lack of an aetiologically based nosology classification has contributed to instability in psychiatric diagnoses over time. This study aimed to examine the diagnostic stability of psychosis diagnoses using data from an incidence sample of psychosis cases, followed up after 10 years and to examine those baseline variables which were associated with diagnostic change. METHOD: Data were examined from the ÆSOP and ÆSOP-10 studies, an incidence and follow-up study, respectively, of a population-based cohort of first-episode psychosis cases from two sites. Diagnosis was assigned using ICD-10 and DSM-IV-TR. Diagnostic change was examined using prospective and retrospective consistency. Baseline variables associated with change were examined using logistic regression and likelihood ratio tests. RESULTS: Slightly more (59.6%) cases had the same baseline and lifetime ICD-10 diagnosis compared with DSM-IV-TR (55.3%), but prospective and retrospective consistency was similar. Schizophrenia, psychotic bipolar disorder and drug-induced psychosis were more prospectively consistent than other diagnoses. A substantial number of cases with other diagnoses at baseline (ICD-10, n = 61; DSM-IV-TR, n = 76) were classified as having schizophrenia at 10 years. Many variables were associated with change to schizophrenia but few with overall change in diagnosis. CONCLUSIONS: Diagnoses other than schizophrenia should to be regarded as potentially provisional.