Artificial neural network (ANN) has been shown its superior predictive power compared with the conventional approaches in many studies. However, they always be treated as ‘black box’ because it provides little explanatory insight into the relative influence of the independent variables in the prediction process. Ghajar et. al. (2002) used the ANN method to develop an empirical correlation for the heat transfer data for a horizontal tube with a reentrant inlet under uniform wall heat flux boundary condition. In their work, the least and the most important variables were examined using an index of importance obtained from the weight and bias matrices based on a single training. However, the method is only applied to one set of experimental data and whether this method can be applied to other sets of data is still not known. Therefore, in this study, additional experimental data for different inlet configurations, namely, the square-edge and bell-mouth inlets from Ghajar and Tarn (1994) in the turbulent region are used to further verify this method. Unlike the previous research, the index of importance in this study is from the average of multiple trainings. Furthermore, the number of neurons and the computation iteration are also changed. Using the methodology, the Reynolds number (Re) can be observed as the most important parameters and the length-to-diameter ratio (X/D) is the least important parameter except for the bell-mouth inlet. The findings agree well with the physical meanings for convective heat transfer in the turbulent flow region.