When performing localization on such a large scale, the computation of the 2D-3D matches quickly becomes the bottleneck of the localization pipeline. Therefore, efficient and effective algorithms for correspondence search are required. In this talk, we look at the two dominant approaches for 2D-3D matching: Direct search, which directly estimates correspondences between the image and the model, and indirect matching, which uses intermediate structures to limit the search space. We compare both approaches and discuss their advantages and disadvantages. Based on the example of state-of-the-art methods, we highlight the steps crucial for good performance for both direct and indirect search.