Abstract: Alignment in cryogenic electron tomography(CryoET) is an important processing step for biomolecules three-dimensional reconstruction. Alignment accuracy is critical to 3D reconstruction quality. Marker-free alignment has the advantages of wide applicability, not selecting markers, and no contamination from markers in the sample. For solving the problem of insufficient utilization of image features and low alignment accuracy in marker-free alignment method, we propose a marker-free alignment method based on affine-invariant feature. More feature points are exploited in the projection images for improving the alignment accuracy. We simulate the rotation by affine transformation and extract SIFT features which are invariant to translation, rotation, scale, and view transformations. In this process, characteristics of cryo-electron tomography imaging are fully considered to reduce the dimensionality of the simulation space for greatly improving the computational efficiency. Then, a large number of mismatches are removed through motion consistency constraints, and correct feature matches are obtained through the RANSAC method in the second screening, obtaining the alignment parameters of the cryo-electron tomography images. Experiments on the public CryoET datasets indicate that our method effectively increases the number of retained feature points, reduces the reprojection error, improves the accuracy of alignment while improving the efficiency of computing affine invariant features.