A method based on improved affinity propagation clustering of data warehouse is proposed by analyzing physicians' antibiotic prescribing practices in this article. Loading pivottable by the data warehouse, we can select a representative physicians' antibiotic prescription data and find the internal factors of antibiotic prescribing behavior of physicians through improved affinity propagation. The data is from the antibiotic prescription of information system of a first-class hospital in Zhejiang province in 2012, which is studied through cross-sectional. Firstly, we build the data warehouse, then cluster and reduce the dimensionality of data set by improved affinity propagation clustering to get the training set and test set. Finally, the results show that different departments, months and types of antibiotics have significant impacts on the data of antibiotic prescriptions. There are significant differences of cephalosporin and penicillin with seasonal changes in this hospital which can be used as evaluation index of the doctor's prescription.