Dermoscopy is the examination of the skin using skin surface microscopy. We develop a novel method for classifying melanocytic tumors as benign or malignant by the analysis of digital dermoscopy images. In this paper before classification pre-processing and segmentation, feature extraction is to be carried out. Then introduce the classifier based on neural network ensemble is used to classify the melanocytic tumors as benign or malignant. To improve the neural network's generalization performance proposes an effective neural network ensemble approach with an idea. One is to apply neural network's output sensitivity as a measure to evaluate neural network's output diversity at the input near training samples so as to able to select a diverse individual from a pool of well-trained neural network. Lesions are extracted using SGNN. Lesions are classified using a classifier based on neural network ensemble model.