If documents are all the same length, BM15 is the same as BM11. The fact that BM11 was so much better than BM15 must reflect the very large variation in document length in the test collection. BM15 without a document length correction is not very much better than BM1 (Table 1, rows 2 and 3). For this reason, experiments are currently under way searching relatively short, uniform documents. Clearly, statistical characteristics of different databases (e.g. abstracts, controlled and/or free indexing) will vary widely in ways which may substantially affect the performance of the weighting functions discussed here.