Text mining for biomedicine requires a significant amount of domain knowledge. Much of this information is contained in biomedical ontologies. Developers of text mining applications often look for appropriate ontologies that can be integrated into their systems, rather than develop new ontologies from scratch. However, there is often a lack of documentation of the qualities of the ontologies. A number of methodologies for evaluating ontologies have been developed, but it is difficult for users by using these methods to select an ontology. In this paper, we propose a framework for selecting the most appropriate ontology for a particular text mining application. The framework comprises three components, each of which considers different aspects of requirements of text mining applicationson ontologies. We also present an experiment based on the framework choosing an ontology for a gene normalization system.