BACKGROUND Impairments in the attribution of salience are thought to be fundamental to the development of psychotic symptoms and the onset of psychotic disorders. The aim of the present study was to explore longitudinal alterations in salience processing in ultra-high-risk subjects for psychosis. METHOD A total of 23 ultra-high-risk subjects and 13 healthy controls underwent functional magnetic resonance imaging at two time points (mean interval of 17 months) while performing the Salience Attribution Test to assess neural responses to task-relevant (adaptive salience) and task-irrelevant (aberrant salience) stimulus features. RESULTS At presentation, high-risk subjects were less likely than controls to attribute salience to relevant features, and more likely to attribute salience to irrelevant stimulus features. These behavioural differences were no longer evident at follow-up. When attributing salience to relevant cue features, ultra-high-risk subjects showed less activation than controls in the ventral striatum at both baseline and follow-up. Within the high-risk sample, amelioration of abnormal beliefs over the follow-up period was correlated with an increase in right ventral striatum activation during the attribution of salience to relevant cue features. CONCLUSIONS These findings confirm that salience processing is perturbed in ultra-high-risk subjects for psychosis, that this is linked to alterations in ventral striatum function, and that clinical outcomes are related to longitudinal changes in ventral striatum function during salience processing.