Abstract: Sensitivity must be calibrated when nondestructive electromagnetic testing instrument is used to detect the depth of carburized layer of steel piece. However, due to the limitations of carburization technology and the method of testing the depth of standard carburized layer, it is rather difficult to evaluate the sensitivity testing concerning the depth of thin carburized layer by standard sample calibration. Hence, grey prediction theory is applied to make a study on the sensitivity prediction method based on a limit set of carburized layer depth. It is followed by the prediction of instrument sensitivity after the expansion of virtual depth via data. Experiments indicate that the average precision is above 95% when GM(1.1) Model is applied in testing data modeling of the depth of carburized layer, while the maximum error of prediction is within 5% when GM (1.1) is used in the prediction of the depth of carburized layer.