Identifying units of behavior is a universal problem in ethology. Here we compare three different techniques for identifying natural categories among notes in the chick-a-dee call of the black-capped chickadee (Parus atricapillus). The combinatorial nature of ‘chick-a-dee’ calls, in which different note combinations are presumed to convey different messages, makes a problem of note classification especially relevant. The three techniques we used were (1) multidimensional scaling of inter-note similarities derived from two-dimensional cross-correlation of digital spectrograms, 2) visual sorting of notes by human observers, and 3) k-means cluster analysis based on a principal components analysis of 14 measured acoustic features. The three methods yielded generally concordant results, with closest agreement between the multidimensional scaling and visual classifications. We discuss possible mechanisms of information coding and perception, and the significance of these for the chick-a-dee call system. The techniques we employed provide powerful, complementary tools for the analysis of variation in acoustic communication signals.