There is an ever-growing availability of health data, which in many occasions are recorded in text documents rather than as structured data. Exploiting those data using data mining techniques to try to extract some useful knowledge can provide substantial benefits, but usually the data analysts should not have access to sensitive data. Therefore, the documents should be anonymized before providing them for data analysis or manual tagging prior to applying machine learning techniques. In this paper, we present our experience applying anonymization on unstructured health documents written in Spanish. We have developed a prototype that facilitates the anonymization process, which has been adopted by the Aragon Health Sciences Institute (IACS). Our approach is generic and does not require labeled data, which are expensive and difficult to obtain in this context; instead, we mainly rely on regular expressions, dictionaries of named entities, and the categorization of attributes. An experimental evaluation shows the feasibility and performance of the approach.