The improvements of attention by mindfulness training had been proved. But the effects of mindfulness training on attention to emotional stimuli were mixed. We employed a randomized, controlled design to investigate the effects of mindfulness training on the attention to emotional expressions, and investigated whether baseline levels of dispositional mindfulness and emotional intelligence would moderate the intervention effects. Forty participants received 8-week mindfulness training and another forty participants attended two lectures about mindfulness. All participants completed the visual search task, the Five Facet Mindfulness Questionnaire, and the Emotional Intelligence Scale at both pre-training and post-training. The improvements of search efficiencies were larger in the mindfulness group than those in the control group for sad and angry faces, but not for happy faces (F(2, 156)=4.04, p=0.019, η2=0.05). In addition, baseline emotional intelligence (βsadness=-4.49, t=-2.08, p=0.041, 95%CI [-8.78, -0.19]; βanger=-7.39, t=-4.08, p<0.001, 95%CI [-10.99, -3.78]) but not dispositional mindfulness (βsadness=3.17, t=1.56, p=0.124, 95%CI [-0.89, 7.22]; βanger=-1.89, t=-1.02, p=0.313, 95%CI [-5.61, 1.82]) played a significantly moderating role in the relationship between mindfulness training and emotional attention to sadness and anger. The search efficiencies of negative emotions (i.e., anger and sadness) but not positive emotions (i.e., happiness) were significantly improved by mindfulness training. Individuals with high level of baseline emotional intelligence showed significant improvement on the search efficiencies relative to those with low level of emotional intelligence.