Abstract: Sub-aperture stitching can be used to measure large optical aperture with a relatively smaller aperture interferometer. In the process of sub-aperture testing, if reference surface error is taken into consideration, the stitching result will deviate from truth. As a result, obtaining and compensating reference surface error is necessary. An approach was proposed to obtain and compensate reference surface error by fitting overlap data of each sub-aperture with Zernike terms. A series of Zernike terms were added into traditional stitching function, the form of reference surface that characterized by a Zernike polynomial was obtained by using the least-square method. Experiments had been done on flat and sphere sub-aperture stitching, which were then compared with QED's result. The deviation of PV and RMS of these two results are less than 5 nm and 0.2 nm respectively. At the same time, the mismatch of stitching aperture is less than 10 nm. Experimental result proves that reference surface error can be compensated in the process of sub-aperture stitching and stitching aperture is more accurate.