This paper studies the consensus of multiagent systems with binary-valued communication. We consider a group of agents on an undirected graph with a fixed topology, differing from most existing work, the information each agent gets from its neighbors are binary-valued measurements with stochastic noises. A two-time-scale protocol that alternates estimation and control is constructed: Each agent in the system estimates the states of its neighbors' based on the binary-valued measurements for a dwelt time, during which every agent's state will keep constant; Based on the estimations of its neighbors' states and its own state, each agent designs its own control, by which the state will be updated and then repeats the process of estimation alternating with control. For the state estimation, we propose an identification algorithm without truncation, which is analyzed to be convergent and the convergence rate is given. By the two-time-scale protocol, the system is shown to achieve weak consensus and mean square consensus by using a stochastic Lyapunov analysis and the consensus speed is given. Simulations are given with different holding time, which are consistent with the theoretical results.