In multimedia information processing, while the previous focus was on image/video retrieval, content-based categorization and retrieval of 3D computer graphics model is becoming increasingly important. This is due to the increased adoption of 3D graphics representations in multimedia applications and the resulting need for rapid virtual scene assembly from a repository of 3D models. Motivated by these requirements, the main focus of this paper is on the content-based classification and retrieval of 3D computer graphics models based on a histogram feature representation, and the search for an adaptive transformation of this representation such that the resulting classification and retrieval accuracies are optimized. Observing that a histogram is basically an approximation of the probability density function of an underlying random variable, and that a suitable transformation, when applied to the random variable, will allow the classifier to attain better accuracy based on this new representation, we propose an evolutionary optimization approach to search for this set of optimal transformations due to the large size of the search space. In particular, we consider the special class of transformations that take the form of a piecewise continuous mapping. In this case, the transformed variable is a mixed random variable, with both discrete and continuous components, which provides added flexibility for modeling a number of more diverse random variable types. With a suitably defined fitness function for evolutionary strategies (ES) that measures the capability of the transformed histogram representation to induce the correct class structure, our proposed approach is capable of improving the head model classification performance, which in turn allows, in the case of content-based retrieval, the correct preassignment of a query object to its correct class for more efficient search, even in those cases where the query is ambiguous and difficult to characterize.