This project was carried out under the responsability of the Strategy-Economics-Program Divison of the Institut Francais du Petrole. The goal was the short-term (12 months) forecasting of the demand with regard to the four leading petroleum products in France - gas oil (GO), automotive fuel (CA), home heating oil (FOD) and heavy fuel oil (FL). It was decided to test an original method [1]) and to compare it with the widely used Box and Jenkins method [2], which gives good results for the GO and CA series but which proves disappointing for the FOD and FL series. This study is in two parts: (1) the first part describes the original method by breaking it down into trends and seasonality, with the model being additive or multiplicative. We improved its performances by using the theory of the Weiner filter; (2) the second part concerns Box and Jenkins modeling. This model was used on the unprocessed series and then on the series corrected for the influence of working days with the help of the ＇＇Census-X11＇＇ deseasonalization program. For each method, the principal phases are described for the modeling of gas oil on the French domestic market. For the other products, only the principal results are given, i.e. structure of the model retained, matching with reality, reliability of forecasts.