Dynamic variation in power system frequency is required to be estimated in order to implement correcting measures. This study presents power system frequency estimation by using a leaky least mean-square algorithm. The proposed approach uses a variable leak adjustment technique to avoid drifting of the weights involved in the estimation mechanism. A variable adaptation step size is also incorporated in the algorithm to attain faster convergence. The performance of the proposed algorithm is studied through simulations for several critical cases that often arise in a power system. These studies show that the present algorithm is superior over the existing ones in estimating power system frequency. Studies made on experimental data also support the superiority.