To develop statistical models for predicting weight loss and regain, we analyzed the phenotypic responses in an outpatient study of 60 obese subjects randomized to one of three 12-week interventions, diet (-600 kcal) alone, diet with exercise, and diet with sibutramine. This was followed by 12 weeks of observation. The best of the "baseline covariates" models was one that incorporated intervention group and baseline homeostasis model assessment-estimated insulin resistance (HOMA(IR)). It predicted week 12 weight change with R(2) of 0.38 and root mean square error (√MSE) of 2.92 kg. An alternative model incorporating baseline fat mass plus change in weight and HOMA(IR) at week 4 improved the prediction (R(2), 0.67, √MSE, 2.19 kg). We could not identify a satisfactory model to predict weight regain. We conclude that prediction of weight loss over 12 weeks is significantly improved when short-term weight change is incorporated into the model. This information could be utilized to forecast the success of a weight-loss program and to motivate and contribute to innovative designing of obesity trials.