In a prospective cohort study of consecutive primary care attenders in south London, the estimated prevalence of non-psychotic psychiatric disorder was 45·6%, using the Revised Clinical Interview Schedule (CIS-R). The estimated 12-month incidence rate was 15·7%. Extremely high stability was found in CIS-R scores over 12 months (r = 0·65) and 68·8% of cases at baseline were also cases 12 months later. A clear difference emerged in the types of risk factor which were associated with the incidence and maintenance of disorder in the study population: while socio-economic variables, especially low household income and not having a partner were associated with a worse outcome among prevalent cases at baseline, such variables were only weakly associated with the incidence of psychiatric disorder after adjusting for potential confounders. The latter, notably family psychiatric history and the severity of psychiatric symptoms at baseline were independently associated with the incidence of psychiatric disorder after adjusting for other risk factors, including measures of previous psychiatric disorder.