Abstract: Bottom plate corrosion is the main cause for failure of crude oil tanks. To accurately predict the limit corrosion depth of the in-service tank bottom plate and the residual life of the tank, according to Type II and Type III distribution of extreme value of three parameters based on the extreme statistical theory, this paper provides the generalized extreme value (GEV) distribution model, and uses the probability weighting method to estimate GEV model parameters and the moment estimation method to estimate the parameters for Type I extreme value distribution and Type III extreme value distribution of two parameters. The measured data of bottom plate corrosion depth for three groups of tanks are statistically analyzed, and through the test of goodness of fit of the K-S detection and mapping method, the accuracy of different distribution patterns on the corrosion depth prediction are compared. Results show that the generalized extreme value distribution model can well describe the distribution of tank bottom corrosion depth data, but whether it is subject to Type II extreme value distribution or Type III extreme value distribution of three parameters must be determined according to corrosion depth of each tank bottom plate. The GEV model shows the best fitting performance as verified by K-S detection and mapping method.