The wafer positioning mark is characterized by its small size and diverse shapes. Traditional focus evaluation functions exhibit limitations， including poor stability and low sensitivity during computation. To address these challenges， an adaptive focus evaluation algorithm based on multi-directional gradient variance was proposed in this paper. Initially， feature edge pixel points of the mark image were adaptively sampled and extracted. Subsequently， a multi-directional gradient function， tailored for wafer positioning marks， was developed by integrating the Brenner function and the Roberts function. The sampled pixel points were then evaluated to derive the initial focus evaluation function value. Finally， the variance of the initial focus evaluation function values across the image set was computed， thereby enhancing the robustness of the function against noise and yielding the final focus evaluation function value. Experimental results indicate that the proposed algorithm reduces fluctuations in flat areas by 61.49% and enhances sensitivity by a factor of 2.56 compared to traditional evaluation algorithms. This innovative approach provides a superior calculation strategy for achieving high-sensitivity automatic focusing of wafer positioning marks and other micro-nano mark images characterized by distinct edge feature structures.