Abstract: Aiming at the fluctuation effect of the material stress change on the detection signal during the crack defect detection, this paper put forward a signal processing method of normalized derivative to suppress the influence of stress change. Firstly, the single-point detection signal of the tested part was obtained by eddy current detection system, and the mean square deviation of the signal data was caculated to obtain the signal characteristics of single-point acquisition; Then, A-scan detection was carried out by using the obtained signal characteristics to obtain the corresponding scanning signals, and the scanning data were further derived to obtain the mean square deviation derivative signal; Then, the mean square deviation derivative signal was processed by normalization method to obtain normalized derivative signal; Finally, the characteristics of amplitude variation of normalized derivative signals were extracted to evaluate and detect the crack width. The experimental results showed that the normalized derivative signal can effectively restrain the influence of stress change and improve the detection accuracy of crack defects after the detection signal data were processed by normalization method.