Due to the technological advancement in smart buildings and the smart grid, there is increasing desire of managing energy demand in buildings to achieve energy efficiency. In this context, building energy prediction has become an essential approach for measuring building energy performance, assessing energy system efficiency, and developing energy management strategies. In this study, two artificial intelligence techniques (i.e., ANN = artificial neural networks and SVR = support vector regression) are examined and used to predict the peak energy demand to estimate the energy usage for an office building on a university campus based on meteorological and historical energy data. Two-year energy and meteorological data are used, with one year for training and the following year for testing. To investigate the seasonal load trend and the prediction capabilities of the two approaches, two experiments are conducted relying on different scales of training data. In total, 10 prediction models are built, with 8 models implemented on seasonal training datasets and 2 models employed using year-round training data. It is observed that a backpropagation neural network (BPNN) performs better than SVR when dealing with more data, leading to stable generalization and low prediction error. When dealing with less data, it is found that there is no dominance of one approach over another.