A candidate-defect classification method includes the steps of acquiring a scanning electron microscope (SEM) image of a candidate defect detected in an inspection from a sample including a pattern formed thereon, the inspection being preliminarily performed by an other inspection device; computing a feature value of the candidate defect by processing the SEM image; executing defect classification of the candidate defect as any one of a pattern shape defect and an other defect by using the computed feature value; acquiring positional information contained in design data of the pattern with respect to a candidate defect classified as the pattern shape defect; and extracting a systematic defect from among candidate defects classified as the pattern shape defects by performing a comparison of the positional information contained in the design data of the acquired candidate defect to positional information of a portion that has a high probability of causing a pattern formation failure and that has been obtained from the design data of the pattern, the systematic defect being caused due to a layout shape of the pattern, properties of a processor for forming the pattern, or the like.