This decade has seen a great deal of progress in the development of information retrieval systems. Unfortunately, we still lack a systematic understanding of the behavior of the systems and their relationship with documents. In this paper we present a completely new approach towards the understanding of the information retrieval systems. Recently, it has been observed that retrieval systems in TREC 6 show some remarkable patterns in retrieving relevant documents. Based on the TREC 6 observations, we introduce a geometric linear model of information retrieval systems. We then apply the model to predict the number of relevant documents by the retrieval systems. The model is also scalable to a much larger data set. Although the model is developed based on the TREC 6 routing test data, I believe it can be readily applicable to other information retrieval systems. In Appendix, we explained a simple and efficient way of making a better system from the existing systems.