The high performance optical system is generally not easy to align. Particularly, if there are off-axis optical components in the system, the alignment is a very serious problem. Many researchers have reported that using sensitivity of some Zemike coefficients is useful in alignment. However, this method has a weak point in that it generates an accurate result only when the misalignments of component are located within the range where the sensitivities of coefficients are linear to the amount of misalignment. We developed the new method which allows larger misalignment at the early stage of alignment. It is to minimize the merit function of Zernike coefficients rather than to use sensitivities of coefficients. In this paper, we analyzed the cons and pros of conventional alignment method and our own method and demonstrated the accuracy and flexibility of our method by aligning the diameter 900 mm Cassegrain type collimator. Even though there was more than 1 mm decenter error, we could align the secondary mirror to the nominal position with just two trials.