A quantitative structure-property relationship (QSPR) study was performed to predict the melting points of 60 carbocyclic nitroaromatic compounds using the electronic and topologic descriptors computed respectively, with ACD/ ChemSketch and Gaussian 03W programs. The structures of all 60 compounds were optimized using the hybrid density functional theory (DFT) at the B3LYP/6-31G(d) level of theory. In both approaches, 50 compounds were assigned as the training set and the rest as the test set. These compounds were analyzed by the principal components analysis (PCA) method, a descendant multiple linear regression (MLR) analyses and an artificial neural network (ANN). The robustness of the obtained models was assessed by leave-many-out cross-validation, and external validation through test set. This study shows that the PCA and MLR have served also to predict melting point and some other physicochemical properties, but when compared with the results given by the ANN (R=0.997), we realized that the predictions fulfilled by this latter were more effective and much better than other models.