Boosting approaches have been shown in recent years to provide some of the best over-all classification methods. We introduce and analyze LocBoost, a new boosting algorithm.The algorithm is related to models based on incremental maximum-likelihood methods applied to locally weighted mixtures of probabilistic classifiers.We provide upper bounds on the loss of such models in terms of the smoothness properties of the gating functions appearing in the mixture of experts model. Preliminary numerical results appear to be promising. We also discuss conditions under which boosting is guaranteed to learn effectively.