In an alignment apparatus for aligning a mask and a photosensitive substrate (a semiconductor wafer or glass plate applied with a photoresist), and which is suitably used in a projection exposure apparatus (a stepper or aligner), a proximity exposure apparatus, or the like used in a lithography process in the manufacture of a semiconductor element or a liquid crystal display element, two first beams and two second beams differing from the first beams may be radiated on a diffraction grating-like mask mark and a diffraction grating-like substrate mark, respectively, with the two second beams passing through a transparent region adjacent to the mask mark. By detecting diffracted light components of the two first beams and detecting diffracted light components of the two second beams, a relative position shift between the mask and the substrate can be determined. The alignment apparatus advantageously can reduce mixing of alignment light from a mask and alignment light from a wafer (substrate) to a minimum degree, or can sufficiently separate signals corresponding to these light components in a signal processing stage even when mixing inevitably occurs.