For metabolic diseases, functional changes are often earlier than structural lesions, for example, diabetes. The paper aims to provide a survey using Support Vector Machine (SVM) to predict and assess metabolic functions of diabetes based on bio-heat transfer theory and infrared thermal imaging technology. Two metabolic characteristic values, metabolic function parameter and blood perfusion rate, are extracted from thermography data of cold water stimulation experiment as inputs of SVM to set up models by different kernel functions. For more than 2000 clinical data used in the paper, the prediction accuracy averaged 90%. The research provides a new attempt to evaluate diabetes metabolic function, hoping for contribution to early detection of diabetes.