Abstract A QSPR method is presented for predicting the flash point temperature (FPT) of pure compounds in the transportation fuels range. A structural group contribution method is used to determine the flash point temperature using two techniques: multivariable nonlinear regression and artificial neural networks. The method was used to probe the structural groups that have significant contribution to the overall FPT of pure compounds and arrive at the set of 37 atom-type structural groups that can best represent the flash point for about 375 substances. The input parameters to the model are the number of occurrence of each of the 37 structural groups in each molecule. The neural network method was the better of the two techniques and can predict the flash point of pure compounds merely from the knowledge of the molecular structure with an overall correlation coefficient of 0.996 and overall average and maximum errors of 1.12% and 6.62%, respectively. The results are compared to the more traditional approach of the SGC method along with other methods in the literature.