The mental representation and processing of Chinese words are important topics in cognitive psychology. Using ERP technique and priming paradigm, the present study adopted a delayed character-matching task to investigate character-twirling effects in Chinese characters of low frequency, and their interactions with semantic priming. Three characters were presented in succession in each trial of the character-matching task, representing prime, target and probe stimulus respectively. The primes were intact characters, while both the targets and probes were twirled ones. Two variables, priming and twirling, were manipulated with a 2×2 factorial within-subjects design in the experiment, and ended up with 4 treatment conditions, which were primed twirled, primed intact, unprimed twirled, and unprimed intact conditions. The subjects were asked to press a key on keyboard if the target and the probe, which followed the target, were a same character in identity, and press another one if not. Sixteen native Chinese undergraduates or graduates participated in the experiment during which they were instructed to make a yes/no distinction in the task. The behavioral data showed significant interactions between priming and twirled conditions. When the primes and targets were related, twirled targets were even responded to faster than were the intact. Moreover, the usual priming effect and twirled effect on accuracy was obtained, and there was significant interaction. Three-way repeated-measures ANOVAs with factors of Twirling (twirled, intact), Priming (primed, unprimed), and Electrode Site revealed significant main effects of twirling at the 150–300ms interval. The peaks of N170 elicited by the twirled targets were more negative compared with the intact targets. These findings support the notion that N170 reflects a perceptual analysis of visual character configuration. At the 300–500ms interval, similar ANOVAs but conducted on mean amplitude revealed significant main effects of both semantic priming and twirling, and revealed no significant interactions. The amplitude of N400 elicited by target characters was significantly larger in unprimed conditions than in primed trials. More importantly, the amplitude of N400 was significantly larger in twirled than in intact conditions. This N400-like (stimulus-) degradation effect provides an extension to Holcomb (1993), who did not found such effect with English words of similar frequency. This N400-like degradation effect is interpreted in terms of Kutas et al.’s (2006) N400 notion of semantic retrieval and perhaps of Deacon et al.’s (2004) N400 theory of lexical analysis, but it does not support the post-lexical processing theories. The difference in the style of meaning retrieval between Chinese characters and English words was discussed.