Road traffic emission factors (EFs) are important parameters in managing air quality. Estimation typically requires data from advanced (and expensive) monitoring systems which remain unavailable in some regions (e.g. in developing countries). In this context, the use of simpler (lower-cost) systems may be more appropriate, but it is essential to guarantee the robustness of EF estimations. This article describes a methodology designed to estimate vehicle EFs from street canyon measurements of traffic fluxes, wind speed and direction, and pollutant concentration levels by using low-cost devices, all samples at a one-minute interval. We use different moving window filters (time periods) to average the raw measurements. Applying standard multiple linear regressions (MRL) and principal component regressions (PCR), we show that there is an optimal smoothing level that best relates traffic episodes and pollutant concentration measurements. An application for PM10's EFs on four vehicle categories of Havana's fleet shows a preference for PCR over MLR techniques since it reduced the collinearity effects that appear when traffic fluxes are naturally correlated between vehicle categories. The best regression fits (R > 0.5 and standard deviation of estimates < 15%) were obtained by averaging data between 40′ and 60’; within the boundaries of 95% confidence interval motorcycles have an EF = 111.1 ± 2.7 mg km−1 veh−1; modern, light vehicles have an EF = 90.6 ± 11.2 mg km−1 veh−1; old, light vehicles have an EF = 125.4 ± 18.5 mg km−1 veh−1 and heavy vehicles have an EF = 415.1 ± 31.2 mg km−1 veh−1. We showed that upgrading old light vehicles is a promising scenario for reducing PM10 air pollution in Havana by between 10 and 17%.