Abstract: A SVM neural network (support vector machines) for breath sounds recognition algorithm was advanced, breath sounds feature obtained through wavelet analysis were input into neural networks and carried on the training to the training samples as a feature of SVM method input in order to classify the test samples. Three States (normal, mild and severe lesions) of breath sounds were recognized, and K nearest neighbor (KNN) methods are compared . The results show that SVM method has a higher recognition accuracy and can be used to recognize different breath sounds, which settled the local extremum problem that cannot be avoided in the neural network method and provide an effective algorithm for information processing in body area network technology.