After more than 10 years of the U.S. ocean‐bottom seismometer (OBS) Instrument Pool operations, there is still need for a consistent and accurate procedure to determine the orientation of the horizontal seismometer components of passive‐source free‐fall broadband OBSs with respect to geographic north. We present a new Python‐based, automated, and high‐accuracy algorithm to obtain this information during postprocessing of the data. As with some previous methods, our new method Doran–Laske‐Orientation‐Python (DLOPy) is based on measuring intermediate‐period surface‐wave arrival angles from teleseismic earthquakes. A crucial new aspect of DLOPy is the consultation of modern global dispersion maps when setting up the analysis window. We repeat measurements at several frequencies to lower biases from wave propagation in laterally heterogeneous structure. We include measurements from the first minor and major great‐circle arcs to further lower biases caused by uneven geographical data coverage. We demonstrate the high accuracy of our technique through benchmark tests against a well‐established “hands‐on” but slow technique using data from instruments of the Global Seismographic Network for which orientations are well documented. We present results for all Cascadia Initiative deployments, along with a number of other OBS experiments. Compared to other widely used automated codes, DLOPy requires fewer events to achieve the same or better accuracy. This advantage may be greatly beneficial for OBS deployments that last as short as a few months. Our computer code is available for download. It requires minimal user input and is optimized to work with data disseminated through the Incorporated Research Institutions for Seismology Data Management Center.