Decision trees (DTs) are widely used for human dependence assessment in multi-unit (MU) nuclear power plants (NPPs). However, few states of the corresponding influencing factors are provided in DT, and only one state can be selected by experts, which is not confirmed to the actual situation. Thus, the Dempster-Shafer evidence theory (DSET) was used to combine with DT to propose a method for dependence assessment of human failure events (HFEs) in MU-NPP based on uncertainty information fusion. Firstly, the DSET was used to express the uncertain information in experts' judgement, and the basic probability assignment (BPA) was established. Secondly, the group decision result was obtained by fusing the judgment of multiple experts. Finally, the dependence level between HFEs was determined according to DT and fusion judgement, and the joint human error probability could be obtained. The results of case study show that the proposed method can flexibly express and handle uncertainties existing in experts' judgment. It is concluded that the method can achieve more reasonable human reliability analysis (HRA) result.