The work aims to improve the accuracy of printing image registration (IR). A printing image registration method based on gradient and normalized mutual information (NMI) was proposed. The NMI was firstly obtained, and the modulus and direction angle of the image edge gradient were obtained with the edge detection operator. Then, a new measure function was designed based on the edge gradient information and NMI. The Powell was used to optimize the algorithm to obtain the best parameters for registration with the new measure function as the objective function. The proposed method and the NMI-based image registration method were respectively used for printing IR. The results obtained from the statistical experiments showed that, the new method had smaller fluctuation range of registration errors and higher registration accuracy regarding 100 random registration parameters. The proposed method is superior to the NMI-based image registration method in terms of accuracy.