As the existing nonlinear models of mid-long term load forecasting are fairly difficult to apply and their results are not satisfactory， a novel method is presented， which transforms a nonlinear issue into a linear one based on short-term correlation and annual growth. First， linear regression models are constructed in terms of the strong short-term correlation of the preceding year's load. By using a recursive procedure， the weekly average load is then estimated for the next year. Finally， the predicted annual load growth is taken into consideration to modify the estimated values. The validity and practicability of the method proposed are tested with actual data. It is expected that this approach can provide a new feasible solution for mid-long term load forecasting.