To address the learning problems which include unlabeled samples, a novel transfer support vector machine for learning from data with uncertain labels(TSVM-UL) is proposed based on the structure risk minimization model. This method takes knowledge of source domain, the common data between different domains, labeled samples and probabilities of unlabeled samples of target domain into account, and knowledge transferring between the source domain and the target domain is realized. Experiment results on several real-world datasets show the effectiveness of the proposed method.