The extended sparse representation method is robust to image deformation and has shown promising performance in finger vein recognition. However, its large number of dictionary atoms results in high memory usage and long computation time. To address these issues, we propose a direction-guided sparse representation (DGSR) method for finger vein recognition. Our method computes the guidance direction based on rough matching results between the testing and training images, and then selects the corresponding dictionary atoms, reducing the dictionary size and computation time. Additionally, this method uses vein backbones, which are robust to noise, as image features in sparse representation. Experimental results on the open finger vein database from Hong Kong Polytechnic University demonstrate the effectiveness of the proposed method.