Background/Aims: Collaborative care management is effective for improving care of patients with depression and chronic medical conditions. The Care of Mental, Physical, and Substance use Syndromes (COMPASS) project implemented evidence-based core collaborative care management components needed to achieve improvement in patients with both depression (PHQ-9 > 10) and uncontrolled diabetes (hemoglobin A1c > 8.0) or cardiovascular disease (systolic blood pressure > 145 or low-density lipoprotein > 100). Components included measurement-based care, use of a care manager and registry, systematic case reviews, treatment intensification when indicated, relapse/exacerbation prevention and data evaluation for quality improvement. Supported by an award from the Centers for Medicare & Medicaid Services, COMPASS was implemented through a collaborative of eight health systems and over 190 clinics in eight states. We used qualitative data to describe variation in the implementation of core COMPASS components across sites. Methods: Implementation data across the eight COMPASS sites were obtained from annual site visit reports prepared by the Institute for Clinical Systems Improvement and qualitatively analyzed using Atlas.ti software, yielding emergent themes related to implementation facilitators and impediments. The Consolidated Framework for Implementation Research (CFIR) was used to organize qualitative data from site visits and as a conceptual framework for understanding implementation variation across COMPASS sites. CFIR includes six broad categories of implementation: intervention characteristics, outer setting, inner setting, implementation climate, characteristics of individuals, and process. Each broad category contains three to eight implementation subcategories. Results: Four overarching themes were identified from site visit reports: 1) between-site differences; 2) challenges to implementation; 3) COMPASS learning and impact on health system; and 4) staff (characteristics, turnover issues, training, background). Nine additional subthemes were identified that were mapped to CFIR implementation categories and that demonstrated considerable cross-site variation: 1) primary care physician engagement; 2) prior experiences with care coordination; 3) length of patient enrollment in COMPASS model; 4) team dynamics; 5) care manager characteristics; 6) quality improvement reports; 7) registry use; 8) patient social needs; and 9) organizational environment where COMPASS was implemented. Conclusion: Understanding the sources of variation in largescale collaborative care management implementation is critical to increase the odds for further successful dissemination of similar models.