A detected pattern is binarized, the binarized pattern is expanded, an image size is reduced while a connectivity of the expanded pattern is preserved and stored in a first memory. In turn, the binarized pattern is contracted, the image size is reduced while a connectivity of the contracted pattern is preserved and stored in a second memory. Then the expanded pattern is read out from the first memory and a connectivity of the pattern is selected. The contracted pattern is read out from the second pattern and the connectivity of the pattern is extracted. The selected connectivities are compared with the connectivity of a normal pattern to detect a non-coincidence. The circuit pattern having a short circuit or a semi-short circuit defect and a circuit pattern having an open circuit or a semi-open circuit defect are classified and selected in response to these non-coincidences. Further, a pattern shape stored in the first memory is analyzed to specify the position where the short circuit or a semi-short circuit defect is present.