In this article, a set of four-parameter descriptors, ΣMV((subscript ter))R((subscript ter)), L(subscript F), ΔX(subscript SB) and ΣPEI were used to correlate with glass transition temperature T(subscript g) for 84 polymers. Multiple linear regression analysis and back-propagation artificial neural network (ANN) were used to generate the model. The final optimum neural network with 4-8-1 structure produced a training set root mean square error (RMSE) of 3.3k(R^2=0.9975) and a validation set RMSE of 13.9 K(R^2=0.9513). The results show that the ANN model obtained in this paper is accurate in the estimation of T(subscript g) values for polymers.