•An ANN model for predicting fatigue damage induced by wide-band non-Gaussian processes was developed and trained successfully using generated 45,000 samples.•The bandwidth parameters, S-N curve slope, skewness and kurtosis are used as the input neurons of the developed ANN model.•The ANN model shows an excellent agreement with the rainflow counting method. Fatigue damage analysis is essential for ships and offshore structures subjected to various fluctuating loadings. In many realistic scenarios, structural responses that exhibit wide-band non-Gaussian characteristics are common, rendering the frequency-domain methods for fatigue damage analysis inaccurate. The rainflow counting method is widely used in the context of time-domain fatigue damage analysis; however, it entails significant computational cost. In this study, an artificial neural network (ANN) model is developed to predict the fatigue damage induced by wide-band softening and hardening non-Gaussian processes, which is based on a frequency-domain method proposed by Benasciutti and Tovo (2005b). Extensive numerical simulations are conducted based on different power spectra with a broad range of bandwidth parameters, S‒N curve slope, skewness and kurtosis. The corresponding fatigue damage is calculated using the rainflow counting method. The obtained database is used to enhance the generalizability of the ANN model. Case studies are conducted to demonstrate that the developed ANN model can give fatigue damage prediction very close to that obtained from the rainflow counting method.