Clinical decision making is a complex task for physicians as it requires utmost accuracy of diagnosis. This paper proposes a medical decision support system based on least square support vector machine (LS-SVM) and simulated annealing (SA) heuristic for the diagnosis of diabetes in the Pima Indian diabetes (PID) dataset of the UCI machine learning repository. Fisher score (FS) algorithm is used to select the most significant features from the given feature set. LS-SVM with radial basis function (RBF) is used for classification and the SA for optimisation of the kernel parameters of the LS-SVM. The performance of the proposed system is analysed in terms of classification accuracy, sensitivity and specificity using ten-fold cross-validation and confusion matrix. The results show that the classification accuracy of the proposed system outperforms that of various existing systems. © 2016 Inderscience Enterprises Ltd.