Gunnison prairie dogs have been shown to produce alarm calls that incorporate information about predators [Slobodchikoff et al., Anim. Behav. 42, 713-719 (1991)], such as the species of predator and also a description of the color and general size of the predator. The alarm calls contain a series of harmonics that encode this information. The calls have been analyzed with discriminant function analysis (DFA) and classification techniques such as self-organizing neural networks. These tools have been able to show that very specific information is encoded in the calls. However, exactly how this information is encoded into the signal has proven to be elusive. We show examples of the specific information that is encoded, and present a hypothetic model of how that information is encoded into the signals.