People with the same psychiatric disorder can show marked inter-individual variations in clinical outcomes, such as the course of illness or the response to treatment. These differences are not predictable on the basis of a patient’s clinical features, making it difficult to tailor clinical care to their individual needs. However, there is increasing evidence that this clinical heterogeneity reflects underlying neurobiological differences within diagnostic categories. For example, among people at high risk of psychosis, it is difficult to predict who will later develop a psychotic disorder and who will not. However, neuroimaging measures of structure, function, chemistry and connectivity in a hippocampo-striatal-mid brain circuit may be used to differentiate between these two subgroups, and thus permit the targeting of preventive treatment to those who need it most. Similarly, after the onset of psychosis, it is not possible to predict whether a patient will respond to standard treatment or not. However, neuroimaging measures of cingulate glutamate and subcortical dopamine function appear to distinguish between treatment responders and non-responders, and could thus be used to allocate patients to either standard treatment or an alternative. In both these examples, the accuracy of outcome predictions using neuroimaging data may be enhanced by integrating imaging with cognitive, genetic and peripheral blood measures.