This study provides a top-down approach to establish an emission inventory of volatile organic compounds (VOC) based on ambient measurements, by combining the box model and positive matrix factorization (PMF) model. Species-specified VOC emissions, source contributions, and spatial distributions are determined based on regional-scale gridded measurements between September 2008 to December 2009 in the Pearl River Delta (PRD), China. The most prevalent anthropogenic species in the PRD was toluene estimated by the box model to be annual emissions of 167.8 ± 100.5 Gg, followed by m,p-xylene (68.0 ± 45.0 Gg), i-pentane (49.2 ± 40.0 Gg), ethene (47.6 ± 27.6 Gg), n-butane (47.5 ± 40.7 Gg), and benzene (46.8 ± 29.0 Gg). Alkanes such as propane, i-butane, and n-pentane were 2–8 times higher in box model than emission inventories (EI). Species with fewer emissions were highly variable between EI and box model results. Hotspots of VOC emissions were identified in southwestern PRD and port areas, which were not reflected by bottom-up EI. This suggests more research is needed for VOC emissions in the EI, especially for fuel evaporation, industrial operations and marine vessels. The species-specified top-down method can help improve the quality of these emission inventories.