In this paper prediction error measurement (PEM) method is used for online synchronous generator parameters estimation. Unlike the usual off-line standard methods, in the proposed method instead of removing the automatic voltage regulator (AVR) from the circuit and manually disturbing the excitation voltage, a change in the reference signal of the AVR is applied and the parameters are estimated for the AVR-generator interconnected system. To adapt this modeling to real conditions, the input and output signals are mixed with noise. Since the direct application of the PEM method on noisy signals will cause a significant estimation error, the wavelet transform is used as a signal-denoising tool. Using Matlab/Simulink, synchronous generator parameters are estimated under colored noise and white noise conditions. To validate the estimated parameters, the results compare with the standard standstill frequency response (SSFR) test that was applied to the Shahid Rajaee power plant. Moreover, a three-phase short circuit to earth for a period of 2.5 cycles at the generator terminal is simulated and analyzed. The results indicate the proper accuracy of the proposed parameter estimation.