Health information technology has increased accessibility of health and medical data and benefited medical research and healthcare management. However, there are rising concerns about patient privacy in sharing medical and healthcare data. A large amount of these data are in free text form. Existing techniques for privacy-preserving data sharing deal largely with structured data. Current privacy approaches for medical text data focus on detection and removal of patient identifiers from the data, which may be inadequate for protecting privacy or preserving data quality. We propose a new systematic approach to extract, cluster, and anonymize medical text records. Our approach integrates methods developed in both data privacy and health informatics fields. The key novel elements of our approach include a recursive partitioning method to cluster medical text records based on the similarity of the health and medical information and a value-enumeration method to anonymize potentially identifying information in the text data. An experimental study is conducted using real-world medical documents. The results of the experiments demonstrate the effectiveness of the proposed approach.