An autofocusing method is proposed that utilizes cosine score of inner angle between the vectors resulted from vectoring the axial adjacent reconstructed images in digital holography. It is based on the fact that the images near the focus contain more regular features of object than in the defocused region, therefore, the neighboring reconstructed images are more similar to each other at the focus position than defocused and a cosine score is employed to evaluate such similarity. However, the cosine scores between the axial adjacent amplitude images are so close that it is difficult to distinguish the extremum. Therefore, a modified cosine algorithm is presented to offset such problem on consideration of the correlation of the elements, by subtracting the inner product term from the denominator of the cosine algorithm. The cosine and modified cosine score based autofocusing method procedure is first introduced, and then it is utilized in simulation and real holographic data. In simulation, it precisely judges out the actual recording distance and the focus curve shows good focus function criteria, which verifies the method as an ideal circumstance. In real experiment, it can easily search out the focus distance from the focus curve, and it shows good focus judgement ability than most traditional focus metrics selected. Therefore, the feasibility and validation of the proposed autofocusing method are proved by simulation and experiment results.