Characteristics associated with psychiatrist-reported treatment-compliance problems were investigated using the 1999 Study of Psychiatric Patients and Treatments from the Practice Research Network of the American Psychiatric Institute for Research and Education (n=1843). Logistic regression was used to study characteristics associated with compliance problems as perceived by treating psychiatrists. Among the 22 potential predictors of interest, all but three (age, gender, and problems with primary support group) were found to be significantly associated with treatment-compliance problems in bivariate analyses. A predictive model was developed consisting of eight independently significant predictors from diagnostic, clinical, psychosocial, and treatment-history domains. These predictors included substance use disorder diagnosis, medication side effects, moderate to severe psychotic symptoms, personality disorder diagnosis, economic problems, prior hospitalization, current Global Assessment of Functioning scale score, and duration of treatment with current psychiatrist. This predictive model correctly identified the presence or absence of treatment-compliance problems in 91% of patients in a sample randomly drawn from the dataset before model construction. These findings may be useful to clinicians, researchers, and program planners interested in addressing the important issue of treatment-compliance problems in psychiatric care settings.