A fast, physically based, fully nonlinear, and nonsteady reach-averaged river model is developed to provide instantaneous freshwater discharge rates in the tidally influenced stretch of the St. Lawrence River (Canada) from water level measurements alone at two tide gauge stations. The model does not require any a priori knowledge of the river geometry (width and depth) or hydraulic conditions (hydraulic slope and friction coefficient) as those parameters are inferred by inverse modeling. The model is fast in that several decades of hourly discharges can be reproduced in few seconds on any modern desktop computer. The method is accurate in that it reproduces to within 3% the observed tidal discharge variability at 8 different cross-sections distributed along the tidally influenced portion of the St. Lawrence River. Although much simpler, the new method performs as well as, or better than, much more sophisticated models for reproducing daily and monthly discharge averages. Being physically based, it is also shown that the model can perform well outside the limited range of measurements that are generally available for calibration. The method is simple and easy to manage in that the main function only contains a few lines of code such that it could be readily implemented as a tool for real-time discharge monitoring of the St. Lawrence River near Québec City. Being physically based, the model developed here could likely be applied more generally to highly unsteady tidal rivers with flow reversals.