Objective Aim is to predict successful weight loss by metabolic signatures at baseline and to identify which differences in metabolic status may underlie variations in weight loss success. Methods In DiOGenes, a randomized, controlled trial, weight loss was induced using a low-calorie diet (800 kcal) for 8 weeks. Men (N = 236) and women (N = 431) as well as groups with overweight/obesity and morbid obesity were studied separately. The relation between the metabolic status before weight loss and weight loss was assessed by stepwise regression on multiple data sets, including anthropometric parameters, NMR-based plasma metabolites, and LC-MS-based plasma lipid species. Results Maximally, 57% of the variation in weight loss success can be predicted by baseline parameters. The most powerful predictive models were obtained in subjects with morbid obesity. In these models, the metabolites most predictive for weight loss were acetoacetate, triacylglycerols, phosphatidylcholines, specific amino acids, and creatine and creatinine. This metabolic profile suggests that high energy metabolism activity results in higher amounts of weight loss. Conclusions Possible predictive (pre-diet) markers were found for amount of weight loss for specific subgroups.