For the proplem that the training data which have the same distribution with the test data are insuficient, but a lot of training data which have the similar distribution with the test data exist in the related field, a similarity-learning based multi-source transfer learning(SL-MSTL) algorithm is proposed. A similarity-learning based classification model is proposed in contrast to the classical support vector machine(SVM) model. Compared to the SVM model, the proposed similarity-learning based model can make better use of the source information and improve the classification performance. Experimental results show the effectiveness and the practicality of the proposed algorithm.