Localised flux density distributions are closely related to the degraded area around a cut edge or hole. Artificial neural networks (ANN) and MATLAB® Curve Fitting ToolboxTM are useful tools in prediction and analytical calculation of magnetic properties of electrical steels. A 4-node input layer, 1-node output layer model with three hidden layers, and full connectivity between nodes was developed by the ANN for prediction of localised flux density distribution. The input parameters were hole size, cutting method, induction frequency and bulk flux density while the output parameter was localised flux density due to the search coil located at the angles 0º, 25º, 45º and 65º corresponding to the centre of hole and rolling direction. The previous data obtained experimentally was used for training the proposed ANN model. Minimum correlation coefficients and RMS error for the localised flux density were found to be 0.99 and 0.09 respectively after the network was trained. The network was tested using untrained data and then minimum correlation coefficient and RMS error for localised flux density were found to be 0.98 and 0.04 respectively due to test results. A simple analytical equation has also been determined to describe the localised flux distributions. The estimation and calculation results obtained from the ANN model and analytical equations are in good agreement with experimental results previously reported.