Purposes: Mental health treatment dropout is common; however, dropout patterns and predictors are poorly understood in Taiwan. This study explored patterns and predictors of mental health treatment dropout in a nationally representative sample. Methods: Claim data were from the Longitudinal Health Insurance Database 2000 (LHID2000) of Bureau of National Health Insurance. Subjects were diagnostic with schizophrenia and manic-depression psychosis in 2000. Their medical services utilization was analyzed for 1997-2004. Repeated recurrent event survival analyses were used to identify dropout predictors. Results: The main findings of this study are as follows: (1) the female sex, young age, underinsured status, and schizophrenia were associated with high dropout risk. (2) sychiatric comorbidity was associated with a trend toward dropout. (3) Patients who use daycare services for the mentally ill and psychiatric community rehabilitation services were at higher dropout risk than those who did not use these services. (4) Patients who had a catastrophic illness card were at higher dropout risk than those without this card. Conclusions: Study findings underscore the importance of research to understand the clinical significance of dropout predicators, especially for populations at high dropout risk in Taiwan.