BACKGROUND: The many randomized trials of the collaborative care model for improving depression in primary care have not described the implementation and maintenance of this model. This paper reports how and the degree to which collaborative care process changes were implemented and maintained for the 75 primary care clinics participating in the DIAMOND Initiative (Depression Improvement Across Minnesota-Offering a New Direction). METHODS: Each clinic was trained to implement seven components of the model and participated in ongoing evaluation and facilitation activities. For this study, assessment of clinical process implementation was accomplished via completion of surveys by the physician leader and clinic manager of each clinic site at three points in time. The physician leader of each clinic completed a survey measure of the presence of various practice systems prior to and one and two years after implementation. Clinic managers also completed a survey of organizational readiness and the strategies used for implementation. RESULTS: Survey response rates were 96% to 100%. The systems survey confirmed a very high degree of implementation (with large variation) of DIAMOND depression practice systems (mean of 24.4 +/- 14.6%) present at baseline, 57.0 +/- 21.0% at one year (P = <0.0001), and 55.9 +/- 21.3% at two years. There was a similarly large increase (and variation) in the use of various quality improvement strategies for depression (mean of 29.6 +/- 28.1% at baseline, 75.1 +/- 22.3% at one year (P = <0.0001), and 74.6 +/- 23.0% at two years. CONCLUSIONS: This study demonstrates that under the right circumstances, primary care clinics that are prepared to implement evidence-based care can do so if financial barriers are reduced, effective training and facilitation are provided, and the new design introduces the specific mental models, new care processes, and workers and expertise that are needed. Implementation was associated with a marked increase in the number of improvement strategies used, but actual care and outcomes data are needed to associate these changes with patient outcomes and patient-reported care.