OBJECTIVE: To identify psychopathological predictors for suicide in a population of major depressed Diagnostic Statistical Manual-III (DSM-III) in-patients. METHOD: A total of 210 previous participants in multicentre antidepressant drug trials, carried out in a randomized double-blind design, were followed prospectively through a maximum of 10 years. Patients with a drug or alcohol abuse were excluded. The association between suicide and the pretreatment psychopathological profile was analysed using survival statistics. RESULTS: The suicide rate for non-melancholic depressed patients was significantly higher than for melancholic depressed patients. Comorbid personality disorder was independently associated with an increased suicide rate [relative hazard 3.41(CI: 1.15-10.10)]. CONCLUSION: The study indicates that the non-melancholic aspect of depression, and especially comorbid personality disorder, is associated with an increased suicidal vulnerability.