Background. Although lifetime major depression (LTMD) is assessed with only moderate reliability in community samples, some predictors have emerged for ‘reliable’ LTMD. Given the large impact of sex on risk for LTMD, it is of interest to know if there are sex differences in the reliability of LTMD and its predictors.Methods. A total of 5603 members of male–male and male–female twin pairs from a population-based registry were interviewed twice with a mean inter-interview interval of 19 months. LTMD was assessed on each occasion using DSM-III-R criteria. Univariate and multivariate logistic regression analyses were used, combining forward and back-prediction.Results. The long-term test–retest reliability of LTMD was moderate (κ = +0·48) and did not differ significantly between males and females. In a multivariate model, the significant predictors of a stable diagnosis of LTMD, none of which differed across sex, were younger age at onset, older current age, history of treatment, increasing number of symptoms, level of impairment or level of distress, longer duration of episodes, higher current level of depression and the presence during the depressive episode of sad mood, weight loss, hypersomnia or fatigue. Using these variables, it was not possible to predict ‘stably diagnosed’ LTMD with both high sensitivity and high specificity.Conclusion. In community samples, LTMD is diagnosed with moderate reliability. Although diagnostic stability can be predicted by variables related to severity, distress and treatment-seeking (probably acting to make depressive episodes more ‘memorable’), highly accurate prediction of stably diagnosed cases is not possible. Long-term recall is also significantly influenced by current symptoms. Neither the stability of LTMD nor its predictors differ in men and women.