Prior research suggests that the affective priming effect denoting prime‐congruent evaluative judgments about neutral targets preceded by affective primes increases when the primes are processed less deeply. This has been taken as evidence for greater affect misattribution. However, no study so far has combined an experimental manipulation of the depth of prime processing with the benefits of ERPs. Forty‐seven participants made like/dislike responses about Korean ideographs following 800‐ms affective prime words while 64‐channel EEG was recorded. In a randomized within‐subject design, three levels of working‐memory load were applied specifically during prime processing. Affective priming was significant for all loads and even tended to decrease over loads, although efficiency of the load manipulation was confirmed by reduced amplitudes of posterior attention‐sensitive prime ERPs. Moreover, ERPs revealed greater explicit affective discrimination of the prime words as load increased, with strongest valence effects on central/centroparietal N400 and on the parietal/parietooccipital late positive complex under high load. This suggests that (a) participants by default tried to inhibit the processing of the prime's affect, and (b) inhibition more often failed under cognitive load, thus causing emotional breakthrough that resulted in a binding of affect to the prime and, hence, reduced affect misattribution to the target. As a correlate of affective priming in the target ERP, medial‐frontal negativity, a well‐established marker of (low) stimulus value, increased with increasing negative affect of the prime. Findings support implicit prime‐target affect transfer as a major source of affective priming, but also point to the role of strategic top‐down processes.