The present paper presents three different short-term oil models on a weekly basis. With these models we try to forecast oil prices out-of-sample up to three months. Two of the models are based on the VAR methodology and consider fundamental factors like the net long position and oil inventories. The third variant is a pure futures model. It is shown that the first two fundamental models perform better until mid/end 2007 and since mid 2009. During the financial market crisis from end 2007 until mid 2009, the futures model clearly has better forecasting quality than the other models.