Songbird vocal production is a complex nonlinear phenomenon. However, acoustic studies of bird vocalization have mostly been based on linear spectral analysis. Such analysis methods necessarily fail to capture the information content of song, and for that reason are not effective probes of the means by which songbirds communicate. We present a novel approach to the analysis and classification of songbird vocalization using nonlinear time series analysis techniques. Time-delay embedding is used to construct a new coordinate system in which to view the song time series. The number of coordinates required to unfold the dynamics represents the dimensionality of a new geometric space, wherein the song's attractor can be visualized. We show that the reconstructed phase space representation of bird vocalization can reveal information that is absent in traditional linear approaches.