OBJECTIVE: To examine the relationship between patient, disease, and treatment variables and self-reported compliance with asthma controller medications. METHODS: This was a secondary analysis from a cross-sectional study of adults with asthma enrolled in a managed-care organization (MCO). Data were obtained from a mailed questionnaire and the MCO's patient/claims databases. Compliance was reported using the 4-item questionnaire developed by Brooks, et al. and scored as a mean of the responses, with 5 meaning highest compliance. Independent variables included age, gender, race, education, number of comorbidities, years with asthma, health-belief questions, social support, income, number of MDI instructors, inhaler technique, perceived physician access, patient-perceived severity, guideline-derived severity, and three health-related quality of life scores, the Asthma Quality of Life Questionnaire (AQLQ) summary and the physical (PCS) and mental (MCS) component summaries of the SF-36. Multivariate regression analysis was used to determine the independent variables with the strongest relationship to self-reported compliance. Stepwise backward-elimination was used, with the final model consisting of variables with a p <0.05. RESULTS: The 573 respondents were primarily Caucasian (89.5%) and female (71.0%), with an average age of 40.5 ± 12.4 years (mean ± SD) and average asthma duration of 18.3 ± 14.2 years. The mean compliance scale score was 3.7 ± 1.1, with 84.6% indicating some level of noncompliance (score < 5). The final model had an adjusted R2 of 0.26 and included 6 independent variables. Better adherence was associated with longer duration of asthma, more MDI instructors, lack of depression, stronger beliefs in the benefits of treatment and trigger avoidance, and greater perceived severity of asthma. CONCLUSIONS: A complex set of beliefs, perceptions, and experiences constitute the variables associated with compliant medication-taking behavior. Future longitudinal studies should include these variables to determine the predictive strength of the model.