Abstract: In this paper, the downscaling results of the multi-model dataset from the Coupled Model Intercomparison Project Phase 5 (CMIP5) were coupled with the Variable Infiltration Capacity (VIC) model to predict future floods of the Feilaixia reservoir in the Beijiang River basin under RCP2.6, RCP4.5, and RCP8.5 scenarios. Credibility of the projected changes in floods is described using an uncertainty expression approach, as recommended by the Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC). The results indicate that annual maximum flood peak and maximum 7-day and 15-day flood volumes during 2020-2050 would be "about as likely as not" to show an upward trend under the RCP2.6 scenario and "more likely than not" to show an upward trend under the RCP4.5 and RCP8.5 scenarios. Compared with the 1970-2000 historical period, the flood peaks in the 100 a, 50 a, and 20 a return periods are projected to increase under all future scenarios. In contrast, the maximum 7-day and 15-day flood volumes in the 100 a, 50 a, and 20 a return periods would increase under the RCP4.5 scenario and decrease under the RCP2.6 and RCP8.5 scenarios.