A method and system for detecting defects in a physical mask used for fabricating a semiconductor device having multiple layers is disclosed, where each layer has a corresponding mask. The method and system include receiving a digital image of the mask, and automatically detecting edges of the mask in the image using pattern recognition. The detected edges, which are stored in a standard format, are imported along with processing parameters into a process simulator that generates an estimated aerial image of the silicon layout that would be produced by a scanner using the mask and the parameters. The estimated aerial image is then compared to an intended aerial image of the same layer, and any differences found that are greater than predefined tolerances are determined to horizontal defects. In addition, effects that the horizontal defects may have on adjacent layers are analyzed to discover vertical defects.