Treatment-resistant schizophrenia does not exist as a discrete entity, so separating patients who will fail to respond to traditional antipsychotics from those who will respond is impossible with 100% accuracy. However, several predictors of poor clinical outcome have emerged from recent research and knowledge of the processes that lead to poor outcome has become increasingly important with the advent of atypical antipsychotics that may be used in patients with treatment-resistant illness. Much of the variation in outcome can be understood in terms of differences in sample selection, outcome definition and stringency of the diagnostic criteria used. Failure to appreciate these mechanisms may lead to over- or underestimation of the proportion of patients with poor treatment response in clinical and research settings. The importance of factors that predict poor outcome should be judged in terms of their effect size and the degree to which alternative explanations for the association with outcome have been excluded. Although much current research is being focused on specific biological predictors, baseline demographic and illness-related factors, such as ethnic group, sex, social class, type of onset, age of onset and concurrent misuse of alcohol or drugs, have large effects on outcome. Although duration of untreated psychosis before first contact with services may independently predict poor outcome, confounding by variables that are associated with both pathways to care and clinical outcome has not been excluded.