Aiming at the problems of manpower behavior spend and low monitoring accuracy of dairy cows, a cow behavior classification method was proposed based on that firefly algorithm to optimize support vector machine parameters by taking advantage of the data which obtained by wireless transmission neck ring. The method optimized the parameters of the support vector machine by using the firefly optimization algorithm to achieve the optimal classification accuracy. The experimental results showed that the wireless transmission collars can collect and transmit the cow neck activity information simultaneously. And the algorithm could effectively distinguish the three behaviors of different cows’ feeding, ruminating and drinking. The applicability was greatly improved. Among them, the optimal precision, sensitivity and accuracy rate were 97.28%, 97.03% and 98.02%, respectively. Compared with the conventional support vector machine algorithm, using the method proposed, the classification accuracy, sensitivity and accuracy of the same cow were increased by 13.39, 28.2 and 18.8 percentage points, respectively; the classification accuracy, sensitivity and accuracy of different dairy cows were increased by 0.74, 2.24 and 2.12 percentage points, respectively. The research results can provide technical support for further research on abnormal behavior detection and intelligent early warning of diseases in dairy cows.