It is difficult to evaluate the document collection partition in distributed information retrieval. Recently, there is no clear evaluation criterion for the document collection partition problem. In this paper, two partition models are built to formulate the document collection partition problem from the essence of the problem itself and they can be used as the evaluation criterion of the document collection partition problem. A Huffman_encoding_like algorithm is introduced to compute the optimum partition solution given a test query set. The optimum partition solution is a good reference of other partition solution. The experimental results show that the two models are effective document collection partition evaluation criteria.