We were interested in examining the relative effect of using parts of the documents, different combinations of parts of the documents, or whole documents on retrieval and classification. We were also interested in the effect of MeSH terms on retrieval. Our nts show that indexing titles, abstracts, and MeSH terms for adhoc retrieval yielded cally significantly better results than any other part or combination of parts, with abstracts outperforming any other individual part of the documents. In the triage sub-task, using whole s for training a classifier outperformed using titles, abstracts, diagram captions, MeSH terms, and windows of text around gene names. However, training a classifier using the ion of titles, abstracts, and MeSH terms produced results comparable to using whole documents.