This paper presents an analytical method for the optimal estimation of time constants of synchronous machine from Stand Still Frequency Response Testing (SSFR). We show that the analytical method is advantageous over the conventional one since the latter is based on curve fitting representing the variation of the operational inductance as a function of the frequency and provides inaccurate and non-unique solutions. In fact, the analytical method applies the standard theory of linear systems to locate the values of poles and zeros in the frequency response and determines the optimal order of the equivalent circuit that can model the machine accurately. The proposed method is simple, practicable, and effective. However, it needs an optimisation process based on parameter differentiation to improve the values of time constants. Based on the measured data, realistic tests are given to show the advantages of the method.