Land use regression (LUR) has been widely adopted as a method of describing spatial variation in air pollutants; however, traditional LUR methods are not suitable for characterizing short-term or time-variable exposures. Our aim was to develop and validate a spatiotemporal LUR model for use in epidemiological studies examining health effects attributable to time-variable air pollution exposures. A network of 42 NO(2) passive samplers was deployed for 12 two week periods over three years. A mixed effects model was tested using a combination of spatial predictors, and readings from fixed site continuous monitors, in order to predict NO(2) values for any two week period over three years in the defined study area. The final model, including terms based on traffic density at 50 and 150 m, population density within 500 m, commercial land use area within 750 m, and NO(2) concentrations at a central fixed site monitor, explained over 80% of the overall variation in NO(2) concentrations. We suggest that such a model can be used to study the association between variable air pollutant exposures and health effects in epidemiological studies.