In view of the risk measure of RMB/USD exchange rate, this paper proposes a method to measure the exchange rate risk by quantile regression. Firstly, by using the data of the central parity rate of the RMB /USD from August 11, 2015 to September 16, 2019, we choose EGARCH model and TGARCH model to describe the asymmetry, volatility clustering and peak thick tail characteristics of foreign exchange rate series. Next, on the basis of GARCH-type VaR models, we establish QR-GARCH-type VaR models. Finally, Kupiec failure rate test and dynamic quantile test as the post test method are selected to compare the risk prediction accuracy of the two kinds of models. The results show that compared with GARCH-type VaR models, QR-GARCH-type VaR models are not sensitive to the hypothetical distribution of random disturbances and show significantly excellent risk prediction ability. Among them, QR-EGARCH VaR models based on t distribution have the best prediction ability, so QR-GARCH-type VaR models are more applicable and robust in the measurement of the risk of RMB/USD exchange rate.