This paper presents a selective order autoregressive model to forecast electricity demand. In the first stage, the autoregressive model is developed by critically selecting the number of lags from historical demand including to the forecasting model. At this stage, the change in the model performance is recorded in terms of the number of lags which result in optimum performance of the forecasting model. In the next stage, different seasonal patterns will be carefully selected and added to the autoregressive model to achieve further improvements. Finally, the selective order autoregressive model will be introduced by eliminating the insignificant lags from the seasonal autoregressive model. With the aid of the demand data from two different sources namely the state of New South Wales, Australia, and the Global Energy Forecasting Competition 2014, USA, it is demonstrated that the forecasting model successfully includes the seasonality of daily, and weekly periods in the modelling process. Also, it is shown from the obtained results that the orders of seasonality should be changed for different datasets in view of the uniqueness associated with different geographical locations.