In this paper we propose improved exact distributed recursive least-squares algorithm, (dRLS) for decentralized estimation over sensor networks. We consider a situation where there are a number of sensors with high observation noise (‘noisy sensors’) in the network. To deal with noisy sensors, first we show that when there are such sensors in the network, the performance of incremental dRLS algorithm drastically decreases; In addition, by detecting and ignoring these sensors better performance in a sense of estimation can be achieved. To address the problem of noisy sensors, we propose a new algorithm which consists of noisy sensors detection method and modified dRLS algorithm. As our simulation results show, the proposed method outperforms the dRLS algorithm in the same condition.