How insight, paranoia and depression evolve in relation to each other during and after the first episode of schizophrenia is poorly understood but of clinical importance. Serial assessments over 18 months were made using multiple instruments in a consecutive sample of 257 patients with first episode DSM-IV non-affective psychosis. Repeated measures of paranoia, insight, depression and self-esteem were analysed using structural equation modelling, to examine the direction of relationships over time after controlling for confounds. Depression was predicted directly by greater insight, particularly at baseline, and by greater paranoia at every stage of follow-up. Neither relationship was mediated by self-esteem, although there was a weak association of lower self-esteem with greater depression and better insight. Paranoia was not strongly associated with insight. Duration of untreated psychosis and substance use at baseline predicted depression at 18 months. In first-episode psychosis, good insight predicts depression. Subsequently, paranoia is the strongest predictor. Neither effect is mediated by low self-esteem. Effective treatment of positive symptoms is important in preventing and treating low mood in early schizophrenia.