To improve the prediction accuracy of unstable electric load time series, a prediction method was proposed based on the decomposition of Haar wavelet and the coordination of ARIMAX model. First, the high and low frequency information sequences were separated by the Haar wavelet decomposition from the mixed information sequence, which consists of the informations of electricity price and electric load. Then, the high and low frequency sequences of the electricity price series were used to establish an ARIMAX model for prediction. Finally, the historic load series was obtained by the Harr wavelet reconstructian according to the prediction values, resulted from the ARIMAX analysis, and the results were compared with the traditional time series. The practical use of the method presented shows that it is helpful to increase the prediction accuracy of the instable load time series.