Monthly electric energy demand forecasting plays an important role for the running of power system. China has two tow calendars and they works at the same time. Holidays designed by the lunar calendar affect the regularity of monthly electric load recorded only by the Gregorian one. The normal fuzzy transform is advanced here to quantitatively describe the impact of the Spring Festival and further divided the influence into Jan. and Feb. After excluding the influence, the amended historical data are adopted to training RBF neural network. Experiment results show that because the regularity of raw data is improved, the generalization ability and forecasting precise of RBF neural network are improved.