Using multilevel analysis, this study investigated participant and site characteristics associated with participant retention in a multisite diabetes prevention translational project among American Indian and Alaska Native (AI/AN) people. We analyzed data from the Special Diabetes Program for Indians Diabetes Prevention Program (SDPI-DP), a lifestyle intervention to prevent diabetes implemented in 36 AI/AN grantee sites. A total of 2,553 participants were recruited and started the intervention between January 1, 2006 and July 31, 2008. They were offered the 16-session Lifestyle Balance Curriculum from the Diabetes Prevention Program (DPP) in the first 16-24 weeks of intervention. Generalized estimating equation models and proportional hazards models with robust standard error estimates were used to evaluate the relationships of participant and site characteristics with retention. As of July 31, 2009, about 50 % of SDPI-DP participants were lost to follow-up. Those who were younger, male, with lower household income, no family support person, and more baseline chronic pain were at higher risk for both short-term and long-term retention failure (i.e., not completing all 16 DPP sessions and loss to follow-up, respectively). Sites with large user populations and younger staff had lower likelihood of retaining participants successfully. Other site characteristics related to higher risk for retention failure included staff rating of participant disinterest in SDPI-DP and barriers to participant transportation and child/elder care. Future translational initiatives need to pay attention to both participant- and site-level factors in order to maximize participant retention.