Due to the heterogeneous distribution of discharge–sediment concentration data in most rivers, fitting a single linear sediment rating curve is often inadequate. Consequently, different sediment rating curves are commonly fitted to datasets representing different flow magnitudes. Accordingly, this study aimed to determine the threshold discharge using Bayesian and Genetic Algorithm (GA) approaches and to compare their performance with the median discharge method at the Tamar and Ramian hydrometric stations located in Golestan Province, Iran. After preparing the discharge–sediment datasets, a Bayesian segmented linear regression model was developed and implemented in the WinBUGS software environment. For the Genetic Algorithm approach, the search range for the threshold discharge at both study stations was defined as the interquartile range (25th–75th percentiles) of the observed discharge values. The GA parameters were set to an initial population of 20 chromosomes, a crossover rate of 0.70, and a mutation rate of 0.30. The results demonstrated that, for both calibration and validation datasets, the Genetic Algorithm outperformed the Bayesian and median discharge methods. At the Tamar station, the GA achieved RMSE, Nash–Sutcliffe Efficiency (NSE), and correlation coefficient values of 2443, 0.16, and 0.46, respectively. At the Ramian station, the corresponding values were 1782, 0.35, and 0.77, respectively. Comparison with the Bayesian approach revealed that the threshold discharge estimated by the Genetic Algorithm was located between the 75th and 97.5th percentiles of the observed discharge distribution at the Tamar station, whereas at the Ramian station it was situated between the 2.5th and 25th percentiles.