Early detection of melanoma, the deadliest type of skin cancer, has the potential to reduce morbidity and mortality. This paper presents an automated system for early identification of melanoma, which makes objective judgments based on quantitative measures. The system involves image processing, feature extraction, and support vector machine (SVM) classification. Images of 19 malignant melanoma and 168 benign subjects were collected by a spectroscopic device to reflect morphologies in diseased layers of skin. Features were extracted based on statistical parameters of pixel intensities and were fed into an SVM classifier. The system achieved 92% classification accuracy, 100% sensitivity and 92% specificity.