Fixed-point observation test of circuit patterns using a review SEM can be stably performed while false report is prevented even when variations in the circuit patterns to be observed are large. SEM images sequentially acquired by capturing images of predetermined circuit patterns using a review SEM are stored in a storage means. Images that fit set conditions are selected from the stored SEM images, and are averaged so that an average image (GP image) is created. Pattern test is performed by GP comparison using the GP image. With this, test can be performed while false report is prevented even when variations in the circuit patterns are large.