To develop the three-dimensional recognition system by the multi-matched filter, which recognizes objects by the information from the matched filters with various directions around a 3-D object, the 3-D matched filter is proposed as an elementary technique. Our proposed method uses wavefronts from the reflected lightwave from a 3-D object as the input to the matched filter, instead of the diffracted light in case of the conventional one. It can discriminate between real 3-D object and its image or concave object and convex one. Unshaded objects can also be employed. Simulational experiments are carried out to show the abillity of our method. The correlation output of our method, however, depends on surface condition and positioning of the objects. The influences of the surface roughness and the shift between references and test objects are examined using the comuter simulation. It is explained that the optical coordinate transformation method is effective to reduce the influences of rotation and shift of objects.