For successful assembly of deformable parts, information about their deformation as well as possible misalignments between the holes and their respective mating parts are essential. Such information can be mainly acquired from visual sensors, and mirrors can be incorporated with a camera to obtain the necessary three-dimensional information by stereo vision. In case of visual sensing systems with multiple mirrors, systematic errors need to be reduced by the system calibration and the mirror position adjustment in order to enhance system measurement accuracy. In this paper, a self-imaging method is presented to adjust and calibrate the mirror position of a visual sensing system designed to measure the three-dimensional information in deformable peg-in-hole tasks. It is composed of a CCD camera and a series of mirrors including two pyramidal mirrors. The measurement accuracy of the system is discussed as well. The ideal designed accuracy can be obtained from the design specification, and the actual accuracy can be obtained from experimental results. The designed resolution of the system is about 0.2 mm per pixel of the camera, but actual measurement error is about three times larger than the theoretically possible error according to the designed resolution.