Abstract: A depression detection method based on the multiscale information entropy of eye scanpath is proposed to overcome the disadvantages of high subjective dependence and non-universality in the traditional depression detection. Characterized by the multiscale information entropy of the eye scanpath, this method detected high-risk groups of depression and patients with depression by comparing the differences between the subjects' eye scanpaths under the same semantic stimulus. On the eye movement dataset of issues who answer the self-rating high-risk of depression scale, the average classification accuracy of this method is 80.36%, which has 12.50,11.79, and 9.08 percentage points increase compared with MultiMatch, ScanMatch, and SubsMatch algorithms respectively. Experimental results show that this method can better capture fine-grained eye movement information and has higher accuracy and sensitivity.