An energy‐ and predominant‐period‐dependent (EDP) P‐wave onset automatic picking (EDP‐picker) algorithm is proposed to deal with the problem of inaccurate P‐wave onset picking in cases in which the P‐wave onset is hidden in high‐amplitude ambient noise or the energy difference between the seismic P‐wave and ambient noise is indistinguishable. The algorithm evaluates the energy change using a characteristic variable , which describes the energy increment of the P wave above ambient noise. The period change is evaluated using two variables with respect to the predominant period, namely as proposed by Hildyard et al. (2008) and as the gradient of . The EDP‐picker algorithm has two steps: (1) threshold‐based cursory P‐wave onset picking and (2) precise P‐wave onset picking using an Akaike information criterion function, in which both energy information and period information are considered. All three parameters are determined in a 1 s sliding window. The proposed algorithm is verified on a large dataset comprising 13,481 vertical strong ground motion records for 570 events selected from K‐NET (Japan) and China Strong Motion Networks Center data. For all records with an epicentral distance of less than 150 km, 93.5% of residuals of manual picks and auto picks are within . The results demonstrate that EDP‐picker is robust and suitable for real‐time systems.