Abstract: Consensus analysis and design problem was investigated for linear multi-agent systems with sampled data packet losses in directed communication topology. Random data packet dropouts during operation were described by the Bernoulli-distribution. A proper state linear transformation, constructing from the incidence matrix of a directed spanning tree of the communication topology, was applied to equivalently translate the state consensus problem into an asymptotic stability, one of corresponding systems. By using Lyapunov stability methodology and linear matrix inequality (LMI) techniques, sufficient conditions were derived for assuring all agent's states to achieve asymptotic consensus, and the protocol gains were designed via the stabilizing control method. The relationship among the allowable bounds of sampling period, the control gain matrix, and packet losses probability was analyzed and applied to express the consensus conditions. The effectiveness and feasibility of the proposed approach was verified by simulation examples.