AbstractCalibrated probabilistic forecasts of weekly rainfall were developed for the state of Bihar in northern India and issued in real time during the June–September 2018 monsoon period, up to 2 weeks in advance. The forecasts are based on subseasonal forecasts from the U.S. National Centers for Environmental Prediction CFSv2 model and were calibrated against observed gridded rainfall fields from the India Meteorological Department using canonical correlation analysis. Hindcasts over the 1999–2010 period exhibit appreciable skill at Week 1 lead (Days 3–9), with some skill at Week 2 (Days 10–16), over Bihar as well as over a larger region. Forecasts were issued in real time during the 2018 Indian summer monsoon season for four districts in Bihar on a 1° grid in tercile probability format every Thursday. Verification of the district‐level real‐time forecasts over the 2018 season is evaluated and moderate skill demonstrated in terms of the Brier and Heidke skill scores, especially for the northern districts and for the below‐normal category. Successful monsoon onset and break phase forecasts in 2018 over Bihar are related to episodes of the Madden‐Julian Oscillation, which the model is shown to capture quite well at 1–2 week lead.