Long-term forecasting of water regime characteristics is an important water management problem. In practice, such forecasting with a lead time of one month or more can be carried out within the framework of two probabilistic approaches. The first approach is based on the theory proposed in the works of Box and Jenkins and implemented in the method of autoregressive integrated moving average (ARIMA). The second approach is to apply the learning capabilities of artificial neural networks. The paper presents the results of forecasting average monthly water levels using the example of Lake Ilmen with a lead time from 1 month to 1 year. The comparison of the two methods is carried out, the comparability of the forecasting results is shown.