Proton magnetic resonance spectroscopy at 8.5 T ex vivo was used to investigate skin lesions for metabolic signatures to predict malignancy or indicate malignant potential. Magnetic resonance spectroscopy was performed on biopsy tissue obtained from 63 skin lesions and five melanoma metastases from 55 patients. Samples were grouped and compared according to five clinically significant distinctions: melanoma (n=38) or nonmelanoma (n=30), primary melanoma (n=33) or secondary melanoma (involved nodes and distant metastases, n=5), primary melanoma (n=33) or nevi (n=8), malignant (n=46) or nonmalignant (n=22), and melanocytic (n=46) or nonmelanocytic (n=22). In all comparisons, the average magnetic resonance spectrum of each class lay within 1 standard deviation of the average spectrum of the other class. There was a higher average choline metabolite signal intensity in melanoma-containing biopsies compared with nonmelanoma biopsies. Discriminant analysis based on the intensity of the choline resonance alone achieved 69% accuracy in separation of melanoma and nonmelanoma tissue. Inclusion of other metabolite resonances in the analysis did not increase discrimination accuracy. Tissue heterogeneity in conventionally collected full thickness skin biopsies and possible biochemical variance within individual tissue types limit classification accuracy using the methods and magnetic field strength that were earlier reported to provide accurate discrimination in other cancer types.