Urban water supply systems provide water for a range of water uses from human consumption to fire control, and from garden irrigation to industrial processes. The amount of water consumed in an urban area depends on the activities. Therefore, urban water use is categorised into different sectors such as residential, non-residential and unmetered water use. Modeling urban water use has significant importance in water resources planning and management. From the literature, a vast amount of research was found on modeling residential water use. However, a significant portion (e.g. around 30% of total water use in Melbourne, Australia) of urban water use belongs to nonresidential, and limited attention has been given to modelling non-residential urban water use, possibly due to the difficulty of data collection at end-use level. However, billing data are now available in monthly and quarterly time steps with the latter being widely available. Such data can be utilized to build models for predicting non-residential water use. Therefore, this study aims to develop non-residential water demand models by identifying unique groups based on their homogeneous nature of water use and then further disaggregated into different sub-groups based on their volumes of water uses. Quarterly billing data from 2000 to 2011 were obtained within the Yarra Valley Water (YVW) service area for 46,000 non-residential customers. Of them, monthly water use data available only for 100 customers. Therefore, comparative study of quarterly and monthly time step modeling was carried out by developing a regression model to predict non-residential urban water use in two schools within YVW service area and is presented in this paper. Monthly water use data from 2005 to 2011 were used for this study as water usage patterns are significantly different in pre and post 2005 due in part to permanent water saving rules, which was introduced in 2005. Monthly data were accumulated into quarterly data for developing the quarterly model. Past water use, dummy variables for fixed quarterly effects, incidence of water restrictions, total rainfall and mean monthly maximum daily temperature were considered as influential variables. Although there were five levels of restrictions during this period, due to insufficient data points for some of the restriction levels, all types of restrictions were considered as a single level of restriction. Model performance was measured using Nash-Sutcliffe efficiency (E) and relative error values. Model calibration and validation were performed using two independent data sets by splitting the total record into two periods. Two sets of calibration periods: 2006-2009 and 2006-2010 and validation periods: 2010-2011 and 2011 were also tested to find the best period for model development and validating the models. It was found that the models developed with 2006-2010 calibration period have given better results for both of the schools. Although, both of the monthly and the quarterly time step models were found to predict well, quarterly model performed slightly better than the monthly model with less relative error and good E value.