In this paper we propose an efficient on-line estimation algorithm for determining the size of a dynamic multicast group. By using diffusion approximation and Kalman filter, we derive an estimator that minimizes the mean square of the estimation error. As opposed to previous studies, where the size of the multicast group is supposed to be fixed throughout the estimation procedure, we consider a dynamic estimation scheme that updates the estimation at every observation step. The robustness of our estimator to violation of the assumptions under which it has been derived is addressed via simulations. Further validations of our approach are carried out on real audio traces.