Cooling load forecasting is beneficial for chiller plant operation as it can be used for energy efficient scheduling of chiller plant components (such as chillers, cooling towers etc.). This paper presents a method for forecasting next day's hourly cooling load for a chiller plant using the concepts of similar day selection, wavelet decomposition, and neural networks. Cooling load forecast is obtained from similar day's cooling load by decomposing it into different subbands (frequency components) and training a separate neural network for each component. The accuracy of similar day selection influences the accuracy of the overall method. Hence, four different measures for similar day selection are presented and their results are compared to determine the most accurate measure. The results suggest that the overall method is accurate up to 89% in predicting the cooling load of a chiller plant.