The determination of macromolecular structures requires weighting of experimental evidence relative to prior physical information. Although it can critically affect the quality of the calculated structures, experimental data are routinely weighted on an empirical basis. At present, cross-validation is the most rigorous method to determine the best weight. We describe a general method to adaptively weight experimental data in the course of structure calculation. It is further shown that the necessity to define weights for the data can be completely alleviated. We demonstrate the method on a structure calculation from NMR data and find that the resulting structures are optimal in terms of accuracy and structural quality. Our method is devoid of the bias imposed by an empirical choice of the weight and has some advantages over estimating the weight by cross-validation.Bayesian probability theory ͉ Markov chain Monte Carlo E xperimental data are typically insufficient to determine a biomolecular structure in their own right but need to be complemented with prior physical information. Therefore, structure determination amounts to the search for conformations that have a low physical energy and that, at the same time, minimize a cost function E data quantifying the disagreement between a structural model X and the data. This approach is implemented as minimization of a hybrid energy (1, 2)where the force field E phys compensates a lack of data by imposing physical constraints on the structure. A target function of this form is widely used in macromolecular structure determination, notably from NMR data (3, 4) and from homologyderived restraints (5). The weight w data controls the contribution of the data relative to the force field. Its value can be critical: If it is too large, the contribution of the force field might be too small to avoid overfitting; if the weight is too small, the data contribute too little to define the structure. The choice of the weight also concerns the question of how to judge structural quality. Overfitted structures reach a low R value (6, 7) but exhibit a poor stereochemistry or an unlikely fold. Usually, experimental data are weighted empirically: w data is set ad hoc and held constant during structure calculation. However, already when introducing the hybrid energy concept, Jack and Levitt (1) remarked that correct weighting of the data ''is something of a problem.'' They proposed to adjust the weight to equalize E phys and w data E data ; this adjustment was later refined, for example, in ref. 8. At present, the most rigorous quantitative method to determine the optimal weight is complete crossvalidation (6, 7). However, cross-validation can become unstable and time-consuming in the case of sparse and heterogeneous data with several independent weights.In this work, we introduce an objective and unique way to weight experimental data. We show that a quantitative treatment does not necessitate heuristics like cross-validation: everything we need is contained in the rules of probability ...