The appearance of novel gestural interfaces for instrumental performance, such as virtual character gesture synthesis, yields to a growing interest for the analysis of the expressive components that are at stake during instrumental gestures. Several quantitative studies on instrumental gestures have been carried out, extracting and evaluating gesture parameters for instance based on classification/recognition methods such as k-Means classifiers. Considering gesture synthesis, data-driven techniques for virtual character animation benefit from the realism of human motion capture data. These are based on large sets of gestures in varying contexts, and rely also on features extraction and data reduction methods for keeping the realism of the produced motion and generating style with relatively low computational costs. Our project aims at combining these analysis and synthesis approaches. The present contribution focuses on timpani preparatory gestures, and especially on the evaluation of the relevance of gesture parameters that can be handled by our virtual character animation sytem.