Using VaR-GARCH(1,1) model and four foreign exchange data from 2003 to 2009, this paper studies foreign exchange risk of commercial bank. The study shows that the return of foreign exchange has obvious peak and EUR and JPY meet normal distribution with 1% confidence. According to the estimation and test of GARCH Model, the return of foreign exchange has characteristic of clustering of volatility, which means GARCH(1,1) can be used to simulate the time series of foreign exchange. Under the 99.9% confidence, the maximum potential loss of the foreign exchange portfolio with EUR and JPY estimated by VaR-GARCH(1,1) model is about 0.05% of last trading day's market value.