Abstract: With the rapid development of optoelectronic equipment, the demand for optical testing is increasing, and the large-aperture, high-precision, wide-field-of-view, and high-stability collimators have gradually become key test equipment. A design method for ultra-thin (primary mirror) off-axis split collimator was proposed, which used an ultra-thin primary mirror and a reused air-floating platform as the base of the collimator to reduce the mass of the system. For an ultra-thin primary mirror with a diameter of 1 560 mm and a diameter-to-thickness ratio of 14.8, a 16-point flexible vertical push-pull side support mechanism was designed to eliminate the influence of assembly stress and thermal stress, and a box-type mirror housing and an Invar temperature compensation mechanism were designed to ensure that the distance between the primary and secondary mirrors meets the optical requirements. Based on the Boyes positioning principle, a "three-axis rod + three-tangential rod" composite positioning mechanism and a "V-groove sliding pin + lifting screw" installation mechanism were designed to precisely position the primary mirror. The finite element analysis of the system shows that the RMS (root-mean-square) of gravity deformation of the primary mirror support is 2.546 nm, the support deformation is less than 7.5 nm when the primary mirror support force disturbance is 5 N, the support deformation is less than 5 nm when the primary mirror axial positioning error is ±0.2 mm, the deformation is less than 2.572 nm when the positioning accuracy of the primary mirror support rod is 0.3 mm, and the resonant frequency of the lightweight collimator is greater than 38 Hz. Field tests show that the total weight of the collimator is about 1 800 kg, the horizontal RMS of the primary mirror optical axis is 0.019 5 λ, and the accuracy RMS of the optical tube system is 0.060 λ, all of which are better than the index requirements. Therefore, the proposed off-axis split collimator design method based on the ultra-thin primary mirror is reasonable and feasible, which has a strong reference value for the design of future collimators.