Summary: This paper investigates the problem of information aggregation in a sequential action model. In a model in which agents sequentially take actions after observing the action history and a private signal, an informational cascade is defined as an event in which the sequence of actions converges to a limit and a fully revealing informational cascade is defined as an event in which the sequence converges to a limit which is optimal under the true state. The necessary and sufficient condition for the occurrence of a fully revealing informational cascade almost everywhere is established. The necessary and sufficient condition is interpreted using the characterization by P. R. Milgrom [Econometrica 47, 679-688 (1979; Zbl 0415.90100)] and R. D. McKelvey and T. Page [Econometrica 54, 109-127 (1986; Zbl 0578.90019)].