Bariatric surgery (BS) often leads to remission of type 2 diabetes mellitus (T2DM), but long-term follow-up to determine the durability of T2DM remission and predictors of relapse is needed. We conducted a retrospective cohort study of adults with uncontrolled or medication-controlled T2DM who underwent BS in three health care systems in the HMORN from 1995-2008. A second comparison cohort included all adults with severe obesity (BMI .35) and uncontrolled or medication-controlled T2DM from 2005-2008 (years in which BMI data were available from electronic medical records). Our main outcomes of interest were remission and relapse of T2DM, which were defined using pharmacy and laboratory data. We used multivariable-adjusted Cox proportional hazards models and propensity scores to investigate the impact of BS vs. usual care on diabetes remission. We identified 4,353 adults with uncontrolled or medication-controlled T2DM who had BS from 1995-2008. Most procedures were Roux-en-Y gastric bypass (RYGB; 53.6% open; 42.8% laparoscopic); 1.8% were gastric banding, and 1.8% were other procedures. Overall, 80.0% experienced an initial T2DM remission within five years after BS. Among those with initial T2DM remission, 36.7% relapsed to T2DM within five years. Median duration of T2DM remission was 7.5 years. Significant predictors of remission and relapse were procedure type, poor preoperative glycemic control, insulin use, and longer duration of T2DM. In multivariate analysis of 1,395 BS subjects and 62,167 non-surgical subjects in 2005-2008, BS significantly increased the likelihood of T2DM remission [Hazard Ratio:18.7; 95% Confidence Interval:17.1 to 20.4]. Compared to usual care, BS substantially increases likelihood of T2DM remission, but about one-third of those who achieve remission with BS relapse within 5 years. Caution should be advised in recommending BS as a “cure” for T2DM in severely obese patients.