The forecasting task and some of the main problems, that occur while solving it, were examined in the paper. The main existing forecasting methods, that unfortunately do not take into account these problems, were listed together with their short description. We propose a new approach for building the forecasting methods, which considers some of the mentioned problems. Based on this approach, we constructed a new forecasting method, called ‘linear regression with dynamic weights’, which finds concrete values of weights for the input factors depending on the values of the factors themselves. To test the forecasting abilities of the method we used the set of real time series, for which we built a forecasting model using the proposed method, the “ancestor” method – pure linear regression and the group method of data handling. By analyzing the results we show that the new method produced (on average) better forecasting error than the linear regression, and for some time series its error was better than the group method of data handling produced. In a conclusion we suggested some ways for the future improvement of the method