Identifying individual cells in biomedical images is a challenging problem because of the staining, hardening, and slicing process. A lot of jobs have been done with sufficient staining so the interested cells can be extracted using HIS or other color matching method. When the staining is insufficient, the interested cells will be of the same color with others. Here we introduce a statistical method that can successfully identify leukocyte from others like red cell, intercellular space or megakaryocyte, and thus get the specific count. Learning is required to get the color specific statistical features of interested areas, further identification can be automatic for the whole page of image and the images from the same staining batch. The proposed method can be improved to count the percentage of the leukocytes from a marrow sample and served as an assistant diagnosis of leukemia.