In this work, we present a computerized system to analyze and diagnose melanoma from whole slide H&E stained skin biopsy images. The proposed system first segments epidermis and dermis (E&D) layers by a multi-resolution framework. A set of cytological and textural features are then computed from the segmented E&D layers. The support vector machine (SVM) is finally utilized to classify skin biopsy images into different categories such as melanoma, nevus or normal skin. Experiments on 66 biopsy images indicate that the proposed system achieves over 95% classification accuracies.