The Diabetes Prevention Program (DPP) was a randomized clinical trial of prevention of type 2 diabetes in high-risk people. Troglitazone, an insulin-sensitizing agent, was used initially but was discontinued during the trial. Troglitazone therapy was compared with other DPP interventions, considering both the short-term "in-trial" results and the longer-term results after troglitazone were discontinued. From 1996 to 1998, participants were randomly assigned to treatment with metformin (n = 587), troglitazone (n = 585), double placebo (n = 582), or intensive lifestyle intervention (ILS) (n = 589). Because of concern regarding its liver toxicity, the troglitazone arm was discontinued in June 1998, after which follow-up of all participants continued. During the mean 0.9 year (range 0.5-1.5 years) of troglitazone treatment, the diabetes incidence rate was 3.0 cases/100 person-years, compared with 12.0, 6.7, and 5.1 cases/100 person-years in the placebo, metformin, and ILS participants (P < 0.001, troglitazone vs. placebo; P = 0.02, troglitazone vs. metformin; P = 0.18, troglitazone vs. ILS). This effect of troglitazone was in part due to improved insulin sensitivity with maintenance of insulin secretion. During the 3 years after troglitazone withdrawal, the diabetes incidence rate was almost identical to that of the placebo group. Troglitazone, therefore, markedly reduced the incidence of diabetes during its limited period of use, but this action did not persist. Whether other thiazolidinedione drugs used for longer periods can safely prevent diabetes remains to be determined.