Abstract: Alignment in cryogenic electron tomography (CryoET) is an important processing step for three-dimensional reconstruction of biological macromolecules. Alignment accuracy is critical to 3D reconstruction quality. For solving the problems of insufficient utilization of image features and low alignment accuracy in marker-free alignment methods, a marker-free alignment method is proposed based on affine-invariant feature. The proposed method simulates the rotation via affine transformation, thus reduces the dimensionality of simulation space. It extracts the ASIFT features which are invariant to translation, rotation, scale, and view transformations. Then, a large number of mismatches are removed through the motion consistency constraints and the subsequent RANSAC method. The correct feature matches are obtained. Finally, the alignment parameters of the cryo-electron tomography images are solved. Comparing with the SIFT method and the IMOD software, the proposed method improves computational efficiency by 70%, increases the number of feature point matches by 81% to 172%, thus reduces the reprojection error by 12% to 26%. As a result, it improves the accuracy of CryoET alignment.