Generalized anxiety disorder (GAD) is a common neurosis disorder characterized by sustained nervousness and anxiety with no definite reasons. The processing of attention plays an important role in the pathogenesis and maintenance of anxiety disorder. Some Studies indicate that anxiety individuals have attentional bias towards threat stimuli, which is the key factor of leading to anxiety. Attentional Bias Training (ABT) is a method of using a systematic training task to change this attentional bias. With the help of ABT such as a modified version of the dot-probe task , we can train individuals with anxiety to direct attention away from negative stimuli, which in turn alleviates symptoms of anxiety. Using a modified dot-probe task, the study aims to examine whether the method can modify the negative emotional bias of patients with GAD and relieve their anxiety, meanwhile, analyze the still unclear mechanisms of ABT. Thirty-six patients with GAD were randomly allocated into training group and control group, they need to complete pre- and post- assessment which include emotional assessment and attentional bias assessment, and training task. The attentional bias training comprised a modified dot-probe task where pictures of faces with either a disgust or neutral emotional expression cued different locations on the computer screen. In training group, a probe that participants responded to always followed neutral faces that paired with a disgust face. In the control group, the probe appeared with equal frequency in the location of the disgust and neutral faces. The pictures of faces are real facial expressions selected from the Chinese Facial Affective Picture System. The Positive Affect and Negative Affect scale (PANAS) and State Trait Anxiety Inventory (STAI) were chosen to assess the emotion states of the participants. E-prime program was used to present stimulus and record the reaction time. The statistical data was analyzed with SPSS16.0. The results showed that the negative attentional bias score and negative attentional disengagement score were significantly lower in training group than in control group. It is reflected that the effect of ABT was definitely found in correcting the negative emotional bias of patients with GAD, and this effect obtained is based on reducing the difficulty to disengage attention from negative stimulus. Besides, the post-tested Positive Affect score were higher than the pre-tested score in training group, and the State Anxiety score were lower than the pre-tested in two groups. In summary, we can draw the conclusion that the attentional bias training is effective in modifying the negative emotional bias of patients with GAD and alleviating their anxiety states. Furthermore, it is revealed that these effects are acquired mainly through improving the ability of the patients with GAD to disengage attention from negativity.