Noninvasive estimation of motor unit propagation velocity (MUPV) was reduced to that of the time delay between signals detected by two surface EMG electrodes placed along the muscle fibres. When the cross-correlation function between the signals was used, the problem with temporal resolution arose. Estimation of the time delay in the frequency domain was proposed to overcome this problem. To check whether the cross-correlation and phase-difference methods give the same estimates, the results obtained by both methods were compared through simulation. A different sensitivity of the two methods to the effects of the excitation origin and extinction was found. Besides, the quality of the estimate depended on the electrode arrangement. The longitudinal double difference electrodes were preferable with the phase-difference method, while the MUPV estimates obtained by the cross-correlation technique were more correct when the longitudinal single difference or bipolar transversal double difference electrodes were used. In addition, the estimates obtained by the phase-difference method were more sensitive to the longitudinal scattering of motor end-plates and ends of the fibres, to the fibre lengths and to the negative after-potential magnitude. Such sensitivity could make MUPV estimates incorrect even under a relatively small distance between the motor unit axis and electrode.