A methodology for modeling time series of hourly urban water use is presented based on separating the data into three components: trend, seasonality, and autocorrelation. Each component is represented by a model whose parameters are estimated. The series is transformed by removing each of the three components and the last transformation produces only a random error series. In the process of identifying the most suitable model for the autocorrelation component of the series, a large number of alternative autoregressive moving average (ARMA) models are assessed in terms of statistics that measure accuracy, parsimony, and randomness of the residuals. Hourly spatially aggregated water use in Cincinnati, Ohio, during the month of October, 2008 is modeled as an example. The model explains approximately 50 of the variance of this series, divided as trend (8.28), seasonality (11.51), and autocorrelation (29.94). The methodology will be applied in a future study of a large data set of individual service connection hourly demand series, spatially aggregated under different schemes.