Abstract: The generalized likelihood uncertainty estimation (GLUE) framework has widely been used for uncertainty estimation in hydrologic modeling thanks to its ease of implementation and less strict statistical assumptions about residual errors. However, its subjective factors such as likelihood functions, their threshold values for model classification, and how individual likelihood values are weighted to construct cumulative likelihood distributions play a non-significant role in uncertainty estimation. In this research, we used Bayesian model averaging (BMA), multi-objective optimization, and the k-nearest neighbor (KNN) algorithm within the GLUE framework to replace the conventional likelihood weighting method and compared their performance. We tested two likelihood functions including the Nash-Sutcliffe efficiency (NSE) and flow duration curve (FDC) to evaluate the predictive uncertainty of the Genie Rural (GR) model for the Chehelchay mountain watershed in Minodasht, Golestan province, Iran. The conventional weighting, multi-objective optimization, and KNN methods were more sensitive to the selection of a likelihood function and the FDC likelihood function produces wider predictive uncertainty bounds compared to the NSE function. In contrast, the BMA method produced predictive uncertainty bounds that are more reliable and similar for both likelihood functions, and hence was less sensitive to the selection of a likelihood function. These reliability and insensitivity of a likelihood weighting method to the likelihood function are important features in uncertainty estimation within the GLUE framework.