How to detect weak parts during the structral fatigue test is a difficult problem. Since the general method is inefficient and relies on human experiences， a new method is presented to detect crack based on the data of structral fatigue tests. First， computation efficiency is greatly promoted in merging massive and scattered testing data by applying the technology of load balancing and parallel computing. Second， the faring algorithm based on minimum deformation energy of curves is presented to reduce the noise in the testing dataset. So data support is achieved for the following analysis. Finally， regression model and crack detection criteria are designed， and fatigue crack detection can be accomplished by solving the key parameters in the L-M algorithm. As a result， the detection accuracy is improved. Software system of key structral fatigue testing data in analyzing is developed， and examples show that the proposed method is efficient and reliable.