The paper presents an on-line algorithm for predicting the average temperature of power transformer winding according to its load current. The algorithm has accounted for the whole history of load current and variations of ambient temperature. The overload capabilities of transformer can be increased by monitoring and controlling the winding temperature and the necessary durability of the transformer can be ensured. The algorithm can also forecast the future winding temperature and the time of reaching the max, permitted the assumed future load current and oil temperature.