This article presents a novel neural network‒based approach to the intra‒day forecasting of call arrivals in call centres. We apply the method to individual time series of arrivals for different customer call groups. To train the model, we use historical call data from three months and, for each day, we aggregate the call volume in 288 intervals of 5 minutes. With these data, our method can be used for predicting the call volume in the next 5‒minute interval using either previous real data or previous predictions to iteratively produce multi‒step‒ahead forecasts. We compare our approach with other conventional forecasting techniques. Experimental results provide factual evidence in favour of our approach. Copyright © 2013 John Wiley & Sons, Ltd.