The blood smear analysis has an important role in definite diagnosis of leukemia disease. The WBC's shapes and numbers in a smear area are examined by Hematology experts to diagnose leukemia. The smear stain process and microscope luminance are blinked because of the intense working tempo. At this scheme, unnoticed information about cells can be recovered by an image processing. In this study, at database peripheral smear images which are collected in workday application were segmented by a spatial learning algorithm. This proposed algorithm is composed of markov random filed with k-means and enhancement methods that provides us the segmentation stage truly without luminance and unsuitable stained smear. After segmentation stage, shape and statistical analysis are done every WBC on smear image to get feature vector about Region of Interest. The WBC's are classified at smear by a decision tree algorithm with this feature vector. The classification rate is defined 89%. The results are reported to help the experts.