Based on the day-ahead price research on the United States PJM electricity market from August to November in 2006,this paper proposes a ARIMA, ARCH and combination model of neural network based on the time series to predict the day-ahead price in future 24 hours of United States PJM electricity market, seasonal ARIMA model reflects the price trend, and seasonal, ARCH models reflect the price’s heteroscedasticity, therefore the model can reflect the characteristics of electricity price better, predicted results is excellent and there is broader application prospects.