Since the latter half of 1960s, several supplementary control schemes - PID control, phase compensator, output feedback control etc. - have been studied to damp out the sustained low frequency oscillations in steady state and it is an important problem to select the proper values of the parameters of power system stabilizers (PSS). This paper presents a unified approach to select the optimal parameters of PSSs in multi-machine power system. The performance measure is chosen as a typical quadratic form in time domain and the sensitivity of performance with respect to the PSS parameters is derived using the gradient matrix. To get the sensitivity of performance, it is required to solve two Lyapunov equations whose solutions exist even if the system is unstable. Thus, to get the feasible solution, it must be preceded to check the stability. In this algorithm, it takes a long time to solve Lyapunov equations. To deal with this problem, a new method is also presented and this method contributes to saving of computation time. By considering the actual constraints of parameters, the entire problem is converted to the static minimization problem with some linear constraints and the gradient projection algorithm is applied to solve it. Application to a sample system showed a satisfactory result.