Background: Only one study has used a prospective method to analyze the diagnostic stability of first psychotic episodes in children and adolescents. The Child and Adolescent First-Episode Psychosis Study (CAFEPS) is a 2-year, prospective longitudinal study of early-onset first episodes of psychosis (EO-FEP). Aim: To describe diagnostic stability and the variables related to diagnostic changes. Methods: Participants were 83 patients (aged 9-17 years) with an EO-FEP consecutively attended. They were assessed with a structured interview (Kiddie-Schedule for Affective Disorders and Schizophrenia, Present and Lifetime version) and clinical scales at baseline and after 2 years. Results: The global consistency for all diagnoses was 63.9%. The small group of bipolar disorder had high stability (92.31%) as did the group with schizophrenia spectrum disorders (90.00%). Depressive disorder had lower stability (37.50%) and the lowest values were for psychotic disorder not otherwise specified (11.76%) and brief psychotic disorder (0%).The most frequent diagnostic shift was to schizophrenia spectrum and bipolar disorders. One group of patients did not meet the criteria for any diagnosis at follow-up. Independent predictors of change to schizophrenia spectrum disorders were lower scores on the Children's Global Assessment Scale (CGAS) and the Hamilton Depression Rating Scale. Predictors of not having a diagnosis at follow-up were the CGAS and the Strauss-Carpenter Outcome Scale. Conclusions: Global diagnostic stability was 63.9%. Bipolar and schizophrenia spectrum disorders were the most stable diagnoses, while depressive disorder and other psychosis the least stable. Psychosocial functioning at baseline was a good predictor of diagnosis at follow-up. These data show the need for longitudinal follow-up in EO-FEP before a stable diagnosis is reached.