This paper describes a study design that blends health services and clinical research approaches to examine the cost-effectiveness of treatments and of quality improvement for depression in primary care, managed care practices. Six managed care organizations in Los Angeles (Calif.), San Antonio (Tex.), San Luis Valley (Colo.), Twin Cities (Minn.), and Columbia (Md.) participated. Primary care clinics were randomized to one of two quality improvement interventions or care as usual. Interventions included patient and provider education, nurse-assisted patient assessment, and resources to support appropriate medication management or access to cognitive behavioral therapy. Practices implemented the interventions with study support. Providers and patients selected treatment. Patients with depressive symptoms regardless of comorbidities were eligible. Over 27,000 primary care patients visiting the practices of 181 primary care clinicians were screened for depression, 14% were potentially eligible, and 1356 enrolled into the 2-year longitudinal study. Enrollees were similar to eligibles, but usual care clinic patients tended to be less severely depressed than intervention clinic patients, partly due to clinic staff enthusiasm. The result of the study showed that studying treatment effects and quality improvement in nonacademic settings is feasible, but requires relaxation of design features of experiments that protect internal validity. The trade-off between certainty of causal inference and generalizability to usual care conditions is discussed. The strengths and limitations of this study design are compared to those of clinical trials and recent clinical effectiveness studies.