Typically available nitrogen dioxide ( $${\hbox {NO}}_{2}$$ NO 2 ) measurements are either spatially representative but temporally sparse or temporally representative but spatially sparse. This article illustrates an approach for making use of the strengths of these two types of data to obtain pollution estimates that are representative both spatially and temporally. The model uses three data sources: a Connecticut epidemiological study with average monthly $${\hbox {NO}}_{2}$$ NO 2 measurements at 651 sites in each season for 1 year; hourly measurements at four Environmental Protection Agency (EPA) sites; and, geospatial data on traffic volume, population density, elevation, and land use. We start by establishing a relationship between observations from the epidemiological study and those at the EPA sites at the monthly level. This relationship is assumed to hold for daily $${\hbox {NO}}_{2}$$ NO 2 levels, enabling us to estimate average daily $${\hbox {NO}}_{2}$$ NO 2 levels using EPA data. Validation study confirmed that this assumption is sound and that the method performs best among three alternatives: a linear model, a mixed-effects model, and the proposed mix-effects model that takes into account spatial correlation. The model can provide predictions of daily $${\hbox {NO}}_{2}$$ NO 2 levels at both study sites and random sites that have no measurements at all. Application of this approach provides pollution estimates that make it possible to study short term relationships between $${\hbox {NO}}_{2}$$ NO 2 exposure and respiratory symptoms such as asthma severity. The approach also offers significant cost reduction for future studies of higher temporal/spatial resolution of air pollution levels by making effective use of readily available EPA monitoring site measurements and observations at spatially dispersed sites from epidemiological studies. The model is implemented under the Bayes framework with a Gibbs sampler.