The main aim of the research was to produce the short-term forecasts of traffic loads by means of neural networks (a multilayer perceptron) and traditional linear models such as autoregressive-integrated moving average models (ARIMA) and exponential smoothing. The traffic of a conventional telephone network as well as a packet-switched IP-network has been analysed. The experimental results prove that in most cases the differences in the quality of short-term forecasts produced by neural networks and linear models are not statistically significant. Therefore, under certain circumstances, the application of such complicated and time-consuming methods as neural networks to forecasting real traffic loads can be unreasonable.