In this paper, we conduct a detailed study of the arc-flag approach introduced in [Lau97, Lau04]. Arc-flags are a modification of Dijkstra’s algorithm to accelerate point-to-point (p2p) shortest path computations. The usage of arc-flags avoids exploring unnecessary paths during shortest path query computations. We present two improvements of the original arc-flag method that reduce the pre-calculation times significantly, tweak the efficiency of queries, and cut down the space requirements. First, we improve the preprocessing of the arc-flags by introducing a centralized shortest path algorithm and thereby we overcome the main drawback of the arc-flag method: for the first time it is now possible to apply the pure arc-flag method to large networks such as continental road networks (US network: 24M nodes, 58M edges). Second, we improve the partitioning used in our revised arc-flag method by using multi-way arc separators. Thereby we almost doubled the efficiency of queries on some instances compared to [Lau06] and further reduced the space requirements. This achievement stresses the vital importance of the right choice of partitioning for the performance of the arc-flag method. On the US network, our revised arc-flag method requires only between 4.1 and 15.5 bytes of additional preprocessed data per arc and p2p queries can be answered within 3 to 9 milliseconds (depending on the metric and the underlying partitioning). These results make our revised arc-flags method the currently best-performing purely goal-directed approach. Without having to use combinations with other advanced acceleration techniques or any network compression method we have kept the modification of Dijkstra’s original algorithm minimal—literally only one additional line of code in the route search algorithm needs to be implemented. Its simplicity suggests its usage in existing code bases of route finding applications—whether that be website routers, mobile phones, or personal navigation devices.