To provide a detecting method of defect applicable to automatic detection of the defect of a part having a complicated form by dividing a fluoroscopic image to two classes corresponding to evaluable zone and non- evaluable zone.

The fluoroscopic image of a part is divided into two classes of zones, which correspond homogeneous evaluable zone and non-evaluable zone, respectively, and the result of the division is displayed in the form of a binary image called a mask M. To obtain the mask M, the residual zone having a small size is blocked by utilizing blocking work to the resulting image. When the residual non-evaluable zone is blocked, the resulting image is a two- phase image, and its gray level value shows the evaluable zone and non- evaluable zone of the fluoroscopic image. The boundary of the images is determined so as to provide the binary image, respectively. The binary image constitutes a conditional shape transformation with the mask, and it is applied to the resulting two-phase image so that the shape transformation can clarify a defect.