Muscle fiber conduction velocity (MFCV) is based on the time delay estimation between electromyography recording channels. In order to take into account the variability of the MFCV, we assume that the time delay between the channels varies over time. In the present paper, the Maximum - Likelihood estimation (MLE) of time varying delay for two channels of EMG signals that follow a polynomial model is derived. Monte Carlo simulations are performed at different noise levels in order to evaluate the noise impact of the estimator. The Maximum-Likelihood estimation was achieved by the Newton method. The delay with unknown model (inverse sinusoidal) was also investigated by cutting this delay into many slices. This approach gives the best results by comparison with the other ones.