In recent years, photochemical air pollution has become a significant problem in the São Paulo Metropolitan Area (SPMA). For some air pollutants, especially ozone and particulate matter, concentrations in excess of national air quality standards have been registered. According to data published by the State Environmental Agency (CETESB), approximately 90% of ozone precursors are emitted into the atmosphere by the vehicle fleet [CETESB, 2000. Relatório de Qualidade do ar do Estado de Sao Paulo, 1999; CETESB, 2002. Relatório de Qualidade do ar do Estado de Sao Paulo, 2004]. The estimation of precursor emissions speciation is a rather complex task. Estimating spatial and temporal variation of vehicle emissions is the greatest source of uncertainty. As in other locales, data regarding motor vehicle emissions are scarce. Due to the considerable discrepancies in emission inventories reported in various regions of the world, we evaluated the official emission inventories of non-methane organic gas (NMOG) and nitrogen oxides (NO x ) using an observation-based approach. Ratios of NO x /carbon monoxide (CO) and NMOG/CO were calculated from ambient measurements taken in the early morning (0700–0800) during July and August of 1999. This top-down approach assumes that early morning ambient concentrations of CO, NO x and NMOG are dominated by motor vehicle emissions, and that the photochemical process has not substantially affected the concentrations. Based on these ratios and on the assumption that official inventory of CO emissions is reasonably accurate, on-road motor vehicle emissions of NO x seem to be significantly overestimated and NMOG emissions slightly underestimated. An Eulerian photochemical model, using both the revised motor vehicle emission inventory and the original official emissions provided by CETESB, was applied to an episodic air pollution event in the SPMA (9–12 August 1999). Meteorology fields were obtained from the CALMET model. When the revised, rather than the official, emission inventories are used, ozone and NO concentrations predicted by the California Institute of Technology (CIT) airshed model more closely match observed values.