Many different kinds of shape descriptors have been defined but usually, each of them is only suitable for some particular kinds of shapes. Then, a strategy to improve performance in arbitrary shapes is the use of several descriptors. In this paper, we address the problem of how to combine several shape descriptors into a single representation. We present an adaptation of the boosting algorithm that permits to train a different classifier for each descriptor and combine all these classifiers to obtain a global classifier. The contribution of each descriptor to this final classifier is determined according to its performance along the boosting iterations. Thus, the most relevant descriptors have the greatest influence in the final classifier.