In order to improve the intelligence and dialectical accuracy of the diagnosis of Chinese medicine, we propose an ensemble learning algorithm based on multimodal perturbation strategy, called MPEL algorithm. Firstly, different sample subspaces of the sample domain are generated by multiple sampling. Secondly, different attribute subspaces are separated by  an improved hierarchical clustering feature selection algorithm in the attribute space, and base classifiers with great diversity are trained. Thirdly, the optimal combination of base classifiers is selected through the greedy strategy so that the overall performance of the algorithm is improved. The medical records of Chinese medicine asthma symptomssyndromes are selected to verify the performance of the proposed algorithm. Experimental results show that the proposed algorithm has faster training speed and higher recognition accuracy in the prediction of asthma symptomssyndromes than other current ensemble learning algorithms, and the highest recognition accuracy of the MPEL can reach up to 98.16%.