Abstract: The adsorption performance of volatile organic components (VOCs) on adsorbent was strongly affected by its porous structure. The traditional adsorbent screening required not only porous characterization, but also n-butane working capacity (BWC) measurement. In order to improve the screening efficiency, the average deviation of the experimental BWC and the calculated ones obtained from the BP artificial neural network constructed by machine learning on corresponding characterizations after thirty repeats of 61 kinds of activated carbon samples was about 6.64%. It was significant to explore such quantitative relationship between feature properties of activated carbon and their related BWC, which was meaningful for the further reduction of expenditure on adsorbent screening.