Identification of the roles of different active sites is vital for the rational design of catalysts. We present a cutting-edge strategy to discern the contributions of different single-atom gold species and nanoparticles in 1,3-butadiene hydrogenation, through coupling of advanced spectroscopic techniques, electron microscopy-based automated image analyses, and steady-state and kinetic studies. While all the carbonhosted single gold atoms display negligible initial activity, the in situ-evolved gold nanoparticles are highly active. Full metal-species quantification is realized by combining electron-microscopy-based atom recognition statistics and deep-learning-driven nanoparticle segmentation algorithm, allowing the structure-activity correlations for the hybrid catalysts containing different Au architectures to be established. Surface exposure density of Au nanoparticles, as revealed by electron-microscopy-based statistics, is revealed as a new and reliable activity descriptor.