Although the fulcrum of service provision for personality disorder (PD) has shifted from hospital‐based to psychodynamically‐ and cognitively‐oriented outpatient programmes, very few studies have attempted to compare specialist moderate intensity outpatient programmes with specialist high‐intensity residential models, or to explore whether a period of inpatient treatment may be necessary to improve outcome and prognosis. In this article, we prospectively compare changes over a 4‐year period in 3 groups of patients with personality disorders (N = 162) treated in a specialist community‐based (CBP, N = 30), a step‐down (RT‐CBP, N = 87), and a specialist residential programme (RT, N = 45) in psychiatric distress, deliberate self‐injury, and suicide attempt using multilevel modelling and multivariate logistic regression analyses. The results showed that percentages of early‐dropout were significantly different (p = .0001) for the 3 programmes (CBP = 13.4%, RT‐CBP = 10.2%, and RT = 41.4%). A significant interaction between treatment model and time was found for psychiatric distress (p = .001), with CBP and RT‐CBP achieving more marked changes (g = 1.20 and g = 0.68, respectively) compared to RT (g = 0.30) at 48‐month follow‐up. CBP and RT‐CBP were found to significantly reduce impulsive behaviour (deliberate self‐injury and suicide attempt) compared to RT. Severity of presentation was not found to be a significant predictor of outcome. Long‐term RT showed no advantage over long‐term CBP, either as stand‐alone or as step‐down treatment. Replication may be needed to confirm generalizability of results, and a number of limitations in the study design may moderate the inferences that can be drawn from the results.