This paper deals with output-feedback-based state consensus of homogeneous linear multiagent systems. A novel reduced-order dynamic output-feedback protocol is proposed, which does not need the absolute output information of neighboring agents. Necessary and sufficient conditions for the solvability of the consensus problem under the proposed protocol are derived, and it is shown that the protocol state is convergent if consensus is reached. A design algorithm is then constructed for computing the protocol gains. Furthermore, the protocol specialized for leader-follower consensus is addressed. In addition, the solvability of a truncated version of the proposed protocol without controller interaction is discussed from a low-gain perspective. A numerical example is finally presented to demonstrate the effectiveness of the proposed protocol and design method.