The invention discloses a novel robust generalized high-order cubature kalman filter-based transfer alignment method which comprises the following steps: firstly, in consideration of the nonlinearityof a system, building a transfer alignment nonlinear mathematical model by adopting a velocity, attitude and angular velocity matching method; secondly, performing inertial navigation calculation on amain inertial navigation and a sub inertial navigation, transmitting velocity and attitude information of the main inertial navigation to a navigation computer of the sub inertial navigation, and constructing an observed quantity by using a velocity error, an attitude error and an angular velocity error between the main inertial navigation and the sub inertial navigation; secondly, performing updating to obtain a one-step state forecast value and a covariance based on the generalized high-order cubature kalman filter time, judging a pollution degree of the observed quantity according to a novel Chi-square detection method, and if a judgment result exceeds a preset threshold value, abandoning the observed quantity of the part, and skipping over an observed quantity reconstruction process and a filter updating process; if the judgment result does not exceed the preset threshold value, reconstructing the polluted observed quantity according to a Huber method; finally, performing measurement updating by using the reconstructed observed quantity, and estimating a state value and a covariance of the current moment, so as to realize transfer alignment. The novel robust generalized high-order cubature kalman filter-based transfer alignment method disclosed by the invention solves the problem in quick and high-accuracy alignment of ships under the condition that the observed quantity includes mixed Gaussian noise and outliers.