For the problem of animal emotion recognition, this paper proposes an approach of applying Gaussian Mixture Model algorithm in animal sounds emotion recognition by combining speech processing and machine learning technique. The automatic recognition approach of animal emotion includes three key steps: three feature parameters extraction (Zerocrossing rate, Formant and MelFrequency Cepstral Coefficients), cluster analysis of training samples by using Gaussian Mixture Model, and computation of posterior probability of testing samples. Combination of feature weight coefficients and the number of Gaussian mixture components are analyzed to find the influence to the recognition rate. After that, choosing the optimal parameter, the experiment result shows that the Gaussian Mixture Model algorithm with optimal parameter effectively improves the recognition rate of animal emotion from 84.25% to 96.67%.