TM (Traffic Matrix) estimation is a hot research area recently. Current TM estimation methods are generally designed for backbone and ISP networks. They estimate complete TM which is unnecessary for many IP networks in reality and especially unsuitable for the networks that have many entries. In this paper, we propose an estimation algorithm that is designed for IP networks on link layer. Our algorithm estimates the OD (Origin and Destination pair) count on the basis of link counts which are easy to obtain. Our algorithm first builds a three-entry virtual network from actual network, and then achieves the final result by multivariate linear regression. We verify our algorithm in the official network of our lab by comparing with exact OD count data that are obtained by NetFlow.