When designing or optimizing the aramid honeycomb sandwich panel, it is necessary to predict its sound insulation performance in a wide frequency range (generally 100–5000 Hz). According to the structural characteristics of aramid honeycomb sandwich panels, 49 kinds of aramid honeycomb sandwich panels were designed by the orthogonal test method, and their sound insulation properties were tested. A backpropagation neural network model, a radial basis function neural network model and a general region neural network model are established with the panel material, panel thickness, core material thickness, honeycomb cell diameter, core material density, and acoustic frequency as the input of the artificial neural network model and the sound transmission loss as the output of the artificial neural network model. The results show that the general region neural network model has the highest prediction accuracy, the RMS error value is 3.45%, the prediction effect of the radial basis function neural network is poor, and the RMS error value reaches 11.48. The backpropagation artificial neural network has over fitting in the training process. Thus, it cannot be used to predict the sound insulation performance of the aramid honeycomb sandwich panel under the current data volume.