Abstract: Prediction in ungauged basins is a challenge and hot issue. Parameters regionalization is a useful methodology estimating hydrological model parameters in ungauged basins and has a critical effect on streamflow simulation. With kernel density estimation and Monte Carlo stochastic simulation methods, a framework was constructed to assess the uncertainty of simulated streamflow caused by parameters' error estimated by regionalization methodology. The Xin'anjiang model was applied for streamflow simulation in 42 small and medium-sized catchments with observed hydrologic stations located in the Guangxi Province. As each catchment being supposed an ungauged basin, the parameters of the Xin'anjiang model were calculated by regionalization methodologies including regression-based, similarity-based, and machine learning-based methodology. The performance of flood simulation using regression-based methodology was better than that of the similarity-based methodology. Using optimized machine learning-based regionalization methodology, the flood simulation accuracy was improved by 7%—15%. Compared with calibrated values, there were pronounced errors of model parameters estimated by parameters regionalization methodologies. The errors of sensitive parameters were lower than non-sensitive ones. The results indicated that there were significant uncertainties of randomly modeled floods by Monte Carlo methodology. The relative errors of simulated flood volumes and peak discharges were 10%—30% and 10%—40%, respectively. The results could provide a new technique for streamflow probability modeling and uncertainty assessment in ungauged basins. And this would be useful for flood forecasting and disaster prevention in small and medium-sized rivers.