The distributed estimation problem is one of the most essential issues in sensor networks. This paper studies the consensus estimation problem of linear sensor networks based on the distributed receding horizon estimation (RHE) scheme. To design such a consensus estimation scheme, a novel optimization problem is first formulated for each sensor node, by proposing a new consensus strategy. The explicit solution to each optimization problem is provided and the iterative estimation scheme is established. Under the assumption that the communication graph contains a spanning tree, the sufficient conditions for ensuring robust consensus estimation are developed. It is shown that, the estimation error between any two sensor nodes is upper bounded by a value that is related with the energy bound of the sensor noise.