The conventional expression of instantaneous phase of single frequency interferometer is based on inverse trigonometric functions and consequently is not continuous, which is generally not suitable for analyzing or improving the performance of interferometer synthetically. The feasibility of the Neural Network (NN) approach is investigated by simulations and experiment. Both simulating and experimental results show that the novel continuous model based on NN is applicable to practical, noisy signals, and the precision can be very good. Finally potential applications of the new NN approach, especially to the measurement of velocity and acceleration, are introduced.