The new competitive scenario, that came up as result of the restructuring of the Brazilian Electric Energy Sector, imposes to its agents the need of tools suitable for better resource management. On the specific case of power transformers, which represent one of the most important investment items, the optimal payback involves a suitable balance between revenues related to the energy transported and the actual depreciation costs, mainly those related to the loss of the transformer s useful life, due the degradation of solid insulation by temperature. The present dissertation proposes a time series model, applied to power transformer winding temperature rise forecasting, which combines linear models and artificial neural networks. The main linear forecast methods based on explanatory variables are revised and analyzed, and, together with the proposed model, applied to temperature forecast on real transformers. The results confirm the synergic effect obtained when using linear models with artificial neural networks.