Data mining or data analysis in biomedicine is different from other research fields, because the data in biomedical are heterogeneous and, and they are from different sources. Data from different medical sources are voluminous, each of the resources may have different data structure or data schema, the data quality is also different. Moreover, each physician may have its own interpretation with the same clinical records. In this paper, we analyze the features of medical data, and study data cleaning for medical data in order to mine valuable knowledge. Experiments show that the proposed method is more efficient than the baseline Bayesian network model.