The reduced economic and social activities during the Chinese Spring Festival provide a unique experiment to evaluate reductions in anthropogenic NH3 emissions in China. However, quantifying this unique scenario is challenging as meteorology may mask the real changes in observed NH3 concentrations. Here, we applied a machine learning technique to decouple the effects of meteorology and confirmed that the real (deweathered) NH3 concentration dropped to a minimum during the Spring Festival in 2019 and 2020 at both urban (Beijing) and rural (Xianghe) sites on the North China Plain. Compared with the scenario without the Spring Festival effect, we predicted that NH3 concentrations in 2020 were 39.8% and 24.6% higher than the observed values at the urban and rural sites, respectively. The significant difference between the two sites indicates a larger reduction in anthropogenic NH3 emissions in urban areas than in rural areas due to the Spring Festival and lockdown measures of COVID-19. Future control strategies should consider the emissions of NH3 from the transportation, industrial and residential sectors, considering that agricultural emissions are minor in cold seasons.