The Inherent errors in numerical weather prediction and hydrological models are major sources of uncertainty in watershed runoff forecasting. Bayesian runoff probability forecasting offers an effective approach for forecast correction and uncertainty analysis. This study focuses on the upper reaches of the Three Gorges Basin, using ensemble precipitation forecasts from ECMWF to drive the distributed SWAT hydrological model and produce 1–10 day runoff forecasts. After applying a Box-Cox transformation to normalize the runoff series, a Bayesian Runoff Probability Forecasting (BRPF) model is developed. Two transition probability methods—Interval Transition Probability and Adjacent Transition Probability—are then used to generate probabilistic runoff forecasts for lead times of 1 to 10 days. Results show that the Adjacent Transition Probability method maintains more stable forecasting performance over time, with NSE values exceeding 0.85 at mainstream stations and 0.45 at tributary stations. In contrast, the Interval Transition Probability method exhibits greater variability in forecast accuracy. Both methods enhance runoff forecasting accuracy, but the Adjacent Transition Probability method outperforms the Interval Transition Probability approach.