Long-term electricity load and price forecasts have become critical inputs to energy service provider (ESP) decision makings in restructured environments. This paper presents a three-stage hierarchical approach for long-term electricity load forecasting. These stages are called yearly trend model (YTM), weekly trend model (WTM), and daily trend model (DTM). The first stage fits an appropriate function to data and extracts its yearly trend. The weekly and daily trends are then extracted using the Box-Jenkins method in WTM and DTM, respectively. For doing so, candidate trends are identified using auto correlation function (ACF) and partial auto correlation function (PACF) plots. Then, Akaike information criterion (AIC) and Schwarz information criterion (SIC) are used to select the best-fitted trends. The different behavior of weekends and night times is captured using dummy variables. The obtained yearly, weekly, and daily trends are finally used for electricity load forecasting.