In this paper we propose a rule-based mechanism using Natural Language Processing techniques for extracting biological relations from biomedical text documents. While the rules identify frequently occurring patterns that can be potential relations, significant relations are identified using statistical analysis. Evaluation of the technique has been done on MEDLINE abstracts obtained from the GENIA corpus. Results indicate that our technique has good potential for other text mining applications also. Preliminary analysis shows that indexing biomedical documents on these relations can facilitate high precision document retrieval.