The invention relates to a graphic interface automated test method. The method comprises the following steps of automatically generating random test cases for to-be-tested software, generating records of operation sequences for function modules of test software, and storing a running screenshot; calculating the similarity between the operation sequences according to the records of the operation sequences, and giving out a matching relationship of operations in the operation sequences; clustering the operation sequences according to the similarity between the operation sequences; in each class, arbitrarily selecting one sequence and artificially marking right and wrong; in each class, performing image comparison on other similar sequences and the marked sequence; if comparison results are consistent, directly judging right and wrong, otherwise, regarding the sequence as an undetermined sequence; and if the undetermined sequence exists, returning to perform clustering. According to the method, similar operation sequences are subjected to image comparison in combination with OCR by applying an image comparison algorithm, and it is judged that function defects exist according to results or an iterative process is repeated; and the manual interventions are reduced by using machine learning and image comparison technologies, so that the test automation is realized and the software test efficiency is improved.