INTRODUCTION: Attentional bias modification (ABM) alleviates anxiety by moderating biased attentional processing toward threat; however, its neural mechanisms remain unclear. We examined how ABM changes functional connectivity (FC) and functional network measures, leading to anxiety reduction. METHODS: Fifty-four healthy anxious individuals received either ABM or sham training for 1 month in a double-blind randomized controlled trial. Anxious traits, attentional control, and attentional bias were assessed. Thirty-five participants completed resting-state functional magnetic resonance imaging (MRI) scans before and after training. RESULTS: ABM significantly mitigated an anxious traits regarding physical stress vulnerability (η2 = 0.12, p = 0.009). As compared to sham training, ABM significantly strengthened FC between the pulvinar and transverse temporal gyrus along the temporoparietal junction (T = 3.90, FDR-correctedp = 0.010), whereas it decreased FC between the postCG and ventral fronto-parietal network (vFPN) regions such as the anterior insula and ventrolateral prefrontal cortex (all T ≤ - 3.19, FDR-correctedp ≤ 0.034). Although ABM diminished network measures of the postcentral gyrus (postCG) (all T ≤ - 4.30, FDR-correctedp ≤ 0.006), only the pulvinar-related FC increase was specifically correlated with anxiety reduction (r = - 0.46, p = 0.007). LIMITATIONS: Per-protocol analysis and reduced sample size in MRI analysis. CONCLUSIONS: ABM might augment the pulvinar's control over vFPN to maintain endogenous attention to a behavioral goal, while diminishing the information exchanges of the postCG with vFPN to inhibit the capture of exogenous attention by potential threats. The pulvinar might play a critical role in ABM anxiolytic efficacy.