This paper presents an investigation into the use of occupancy space electrical power demand to mimic occupants' activities in building cooling load prediction by intelligent approach. The occupancy space electrical power demand is obtained from an intelligent networked building power monitoring system. It works as a prototype advanced metering infrastructure - a key feature in smart grid technology. The artificial neural network model adopted is the Levenberg-Marquardt algorithm. The input parameters include the usual external climatic data, hour-type/day-type and pretreated air unit operation schedule, and the occupancy space electrical power demand. The output is the electrical power demand of the building cooling system. Simulation studies are conducted for a university building in Hong Kong. The 2010 and 2011 yearly data is used to conduct simulations. The performance indices used in evaluating the prediction performance are the coefficient of correlation (R), coefficient of variation (CV) and mean absolute percentage error (MAPE). It is demonstrated that with the use of occupancy space electrical power demand as one of the model input parameters, the prediction accuracy of the building cooling load model can be improved. In summer season, the best MAPE and CV is 4.494% and 5.808% respectively for hourly prediction, and is 1.935% and 2.345% respectively for daily prediction. On daily peak cooling load prediction in summer season, the best MAPE and CV is 2.313% and 2.862% respectively. It is found that the variability in the prediction is modest in summer season. (C) 2012 Elsevier B.V. All rights reserved.