The present research applied two types of artificial neural network such as multilayer perceptron and radial basis neural network in order to modeling and predicting audit fees. In this regard, samples include 123 listed firms in the Tehran Stock Exchange through a screening method during March 2013 to March 2022. The research is conducted based on multilayer perceptron and radial artificial neural networks using the MATLAB software. This study used a set of parameters including firm size, current assets to total assets ratio, financial leverage, current assets to current liabilities ratio, long-term liabilities to total assets ratio, quick ratio, loss and financial restatement as input parameters, and audit fees were used as the target parameter in the modeling process. The research results indicate that in training data during March 2013 to March 2019, the radial artificial neural network due to its special structure consistently demonstrates less error compared to multilayer perceptron neural network. But, for model generalization, the performance of the best models was assessed using relevant data during March 2019 to March 2022. In this scenario, the multilayer perceptron neural network was able to predict audit fee with higher accuracy; while the radial artificial neural network failed to predict audit fee accurately. In addition, there is a significant difference between the audit fees predictions using multilayer perceptron neural network compared to radial neural network. Furthermore, the sensitivity analysis results demonstrated that firm size has the most significant impact on the audit fee prediction.