Failure of dams could lead to not only severe economic loss but also human casualties. Breaching analysis is therefore an important component in the risk assessment of dam safety. Traditionally the breaching process is often characterized by the breach parameters including the breach geometry and hydrograph. Over the decades, a large variety of empirical formulas have been proposed on the basis of different dam failure databases to predict the parameters. In most cases, the goodness-of-fit criteria (e.g., minimizing the sum of squares of the fitting errors) remains the only measure of the formula suitability. It may therefore lead to the use of over-complicated formula due to its higher capability of fitting. In this paper, the Bayesian probabilistic approach is used to revisit the problem. The study concludes a set of empirical formulas for the dam breach parameters which are well-balanced between formula fitting capacity and complexity.