This paper presents a multi-step-ahead forecasting method of the electric demand in a large institutional building to be used in the context of demand response control strategy. A cascade-based method is proposed for electric demand forecasting of the cooling system over the next six hours with a time-step of 15 min. Data mining techniques are used for pre-processing the measurements and improving the forecasting models. Data-driven models are developed by using Building Automation System (BAS) trend data of an existing building. First, the air flow rate supplied by the Air Handling Units (AHUs) is forecasted, followed by the cooling coils load, and the whole building cooling load. Finally, the electric demand of the supply fans, chillers and cooling towers, and the total electric demand of the cooling system of the building are forecasted over six hours. The comparison of the forecasted electric demand of the cooling system for the existing building over the six-hour test and the measurements show good agreement with CV(RMSE) of 14.2-22.5%. (C) 2016 Elsevier Ltd. All rights reserved.