In the application of electromagnetic nondestructive evaluation (NDE), the measured signals received by sensors are decided by the flaw associated with operational parameters during inspection. The techniques of invariant pattern recognition have been studied to render NDE signals insensitive to operational variations and preserve or recover crack information. The invariant scheme and algorithms have been addressed to facilitate magnetostatic flux leakage and eddy current NDE, which eliminated operational parameters and lift-off changes corrupt from signal measurements. A novel invariance analysis of eddy current (EC) signals in the inspection of deeply embedded cracks under layered fastener heads has been presented in this paper. A detection system based on uniform EC excitation and giant-magnetoresistive (GMR) pick-up sensors has been developed and shown improved detectability of 2nd and 3rd layer defects around fastener sites in multilayer structures. However, the sensor-tilt due to variation of probe lift-off can generate redundant response as noise effect and obscure the flaw inspection. The variations of GMR sensor-tilt with crack inspection are investigated using an efficient numerical model that simulates the used EC-GMR system. The scheme of invariance transformation is proposed to exclude the sensor-tilt noise and keep the defect inspection identical. The statistical features insensitive to tilt effects are extracted after the invariance processing, which has ensured the probability of flaw detection.