Abstract The violent crime is a highlighted problem of modern society, and it is always having a devastating impact on individuals and their families. While there has been extensive research on the effects of aggressive personality, demographic factors, prior exposure to violence, and psychopathology on violent offenders, there has been a relative dearth of research on the effects of neurocognitive factors on violent offenders. A neurocognitive perspective on violent offenders suggests that violent offenders have signi?cant cognitive impairments that may contribute to their violent lifestyle. Neuropsychological impairments have been reported in male batterers at a higher level than in controls (Teichner et al., 2001). More generally, Eckhardt et al. (1998) observed that batterers tended to commit a variety of cognitive errors, including making assumptions in the absence of objective evidence, and attributing the cause of an event to the hostile intentions of another person. While the source of these cognitive errors is not known, one hypothesis is that violent offenders have a bias towards over-allocating attentional resources to negative affective stimuli, at the expense of appropriate rational processing of ongoing discourse. Then the present study used event-related potentials (ERP) to investigate whether violent offenders are more vigilant to negative emotion. The N2pc ERP component was measured as an index of the allocation of spatial attention. Besides, we want to dig deeper the mechanisms underlie the attentional bias in violent offenders，which provide the basis and method for the prevention of violent crime and correction.. This study tests the hypotheses that (a) violent offenders have a neurocognitive bias favoring negative affect (angry face) stimuli and (b) explore the mechanisms underlie the attentional bias in violent offenders. We recruited 26 male inmates from local prison. 13 violent offenders, mean age 25 years(range 16–31 years, SD=7.07).13 nonviolent offenders, mean age 23 years(range 16–31 years, SD=4.22). Besides, there were no significance difference between violent and nonviolent offenders on state anxiety, and Aggression Questionnaire(P ＞ 0.05). All the participants had self-reported normal or corrected-to-normal eyesight. Three male subjects were excluded from data analysis because of excessive artifacts during EEG recording. Facial stimuli which were made by Facegen Modeller program were used to convey negative information. The faces were comprised by four individuals (2 females) each posing neutral, happy, and angry expressions. The external features of each of the faces were removed and the internal features were presented in a black rectangular frame. The experiment was comprised of five blocks. For each block, 128 trials were delivered. The faces were presented for 500 ms, followed by a random fixation between 500-1000ms. The subjects were told to search the target face(angry or happy face) and response as soon as fast, or for a 1800ms maximum. The inter-trial interval was randomly varied between 500 and 1500 ms. Electroencephalography (EEG) data were recorded from 64 scalp sites using tin electrodes mounted in an elastic cap (brain products). Time windows representing the N2pc component were determined using visual search of individual participant waveforms as has been done in previous research (e.g., Eimer & Kiss, 2007). This ERP component is computed by taking voltage differences between corresponding pairs of electrodes located on the left and right posterior scalp after taking into account the hemi?eld in which attention is deployed .Analyses focused on the correct response trials only and on lateral occipital electrodes (PO7/PO8) within 240-380ms. A two-way repeated measures Analysis of Variance (ANOVA) on the RT and average latencies was conducted, with Group (violent vs. nonviolent offender) as a between-subjects factor, and Target face (happy, angry, neutral) as a within-subjects factors. The ANOVA for RT revealed a signi?cant main effect for Group (violent vs. nonviolent offenders) （F（1，21）=7.997，P＜.010） and Target face ( happy, neural Vs angry face)（F（2，42）=14.573，P＜.01）. Notably, Group×Target face interaction is also signi?cant （F（2，42）=3.492，P＜.05）.Post hoc analysis indicated that RTs were signi?cantly shorter for violent offenders when the target face was “angry’’ relative to “neutral” and ‘‘happy’’. The ANOVA for N2pc response latencies revealed a signi?cant effect for Group×Target face ( F (1，21)=5.446，p=.030), further analysis revealed violent offenders showed significantly shorter response latencies when the target face was “angry’’ relative to “happy’’. There was no significant difference for nonviolent offenders when the target face was “angry’’ and ‘‘happy’’. These findings suggested that violent offenders have attentional bias for negative affect information compare to nonviolent offenders, which have been proved from the behavioral and ERP data. Also these findings suggested that the attentional bias mechanisms of violent offenders tend to be vigilant-avoid.