Platforms for mobile computing and gesture recognitionprovide enticing interfaces for creative expression on virtualmusical instruments. However, sound synthesis on thesesystems is often limited to sample-based synthesizers, whichlimits their expressive capabilities. Source-filter models areadept for such interfaces since they provide flexible, algorithmic sound synthesis, especially in the case of the guitar.In this paper, we present a data-driven approach for modeling guitar excitation signals using principal componentsderived from a corpus of excitation signals. Using thesecomponents as features, we apply nonlinear principal components analysis to derive a feature space that describesthe expressive attributes characteristic to our corpus. Finally, we propose using the reduced dimensionality space asa control interface for an expressive guitar synthesizer.