The curve of relationship between fatigue crack growth rate and the stress strength factor amplitude represented an important fatigue property in designing of damage tolerance limits and predicting life of metallic component parts. In order to have a morereasonable use of testing data, samples from population were stratified suggested by the stratified random sample model (SRAM). The data in each stratum corresponded to the same experiment conditions. A suitable weight was assigned to each stratified sample according to the actual working states of the pressure vessel, so that the estimation of fatigue crack growth rate equation was more accurate for practice. An empirical study shows that the SRAM estimation by using fatigue crack growth rate data from different stoves is obviously better than the estimation from simple random sample model.