Individuals with psychosis are over-represented in the criminal justice system and, as a group, are at elevated risk of re-offending. Recent studies have observed an association between increased contacts with mental health services and reduced re-offending, as well as reduced risk of re-offending in those who are ordered to mental health treatment rather than punitive sanctions. In furthering this work, this study examines the effect of disengagement from mental health treatment on probability of re-offence in offenders with psychosis over time. Data linkage was conducted with judicial, health and mortality datasets from New South Wales, Australia (2001-2015). The study population included 4960 offenders with psychosis who received non-custodial sentences and engaged with community-based mental health treatment. Risk factors for leaving treatment and/or reconviction were examined using multivariate cox regression. Further, a multi-state model was used to observe the probabilities associated with individuals moving between three states: conviction, disengagement from mental health treatment and subsequent re-conviction. A threefold increase was observed in the risk of re-offending for those who disengaged from treatment compared to those who did not (aHR = 2.76, 95% CI 1.65-4.62, p < 0.001). The median time until re-offence was 195days, with the majority (67%) being convicted within one year of leaving treatment. A higher risk of leaving treatment was found for those born outside of Australia, with substance-related psychosis, and a history of violent offence. The findings argue for an emphasis on continued engagement with mental health services following release for offenders with psychosis and identify subgroups within this population for whom concentrated efforts regarding treatment retention should be targeted.