This paper presents a spectral correlation based nonlinear ultrasonic resonance technique for fatigue crack detection. A reduced-order nonlinear oscillator model is initially constructed to illuminate the Contact Acoustic Nonlinearity (CAN) and nonlinear resonance phenomenon. The tailored analytical model considers the rough surface condition of the fatigue cracks, with a crack open-close transition range for the effective modeling of the variable-stiffness CAN. Multiple damage indices (DIs) associated with the degree of nonlinearity of the interrogated materials are then proposed by correlating the ultrasonic resonance spectra. The frequency sweeping signals serve as the excitation waveform to obtain the structural dynamic features. The nonlinear resonance procedure is numerically solved using the central difference method. Short time Fourier transform (STFT) is utilized to extract the resonance spectroscopy. In this study, pristine, linear wave damage interaction case (an open notch case), and nonlinear wave damage interaction case (a fatigue crack case) with various damage severities are considered. Subsequently, three case studies taking advantage of different nonlinear oscillation phenomena are conducted based on the spectral correlation algorithm to detect and monitor the fatigue crack growth: time-history dependence, amplitude dependence, and breakage of superposition. Each of these three nonlinear behaviors can either work individually or collaborate synthetically to detect the nucleation and growth of the fatigue cracks. The proposed nonlinear ultrasonic resonance technique possesses great application potential for fatigue crack detection and quantification. This paper finishes with summary, concluding remarks, and suggestions for future work.