WLAN received signal strength based location estimation techniques are promising for their low cost and ease of deployment. Existing techniques use received signal strength from different access points in an equal way. However, signals sent from the far access point experience more influence of the environmental noises, and more uncertainties are added in the signal strength. As detectable access points have different distance to the user, their contributions to the final estimation should be considered discriminatingly. In this paper, a novel location estimation approach is proposed using the Dempster-Shafer evidence theory to fuse received signal strength from multiple access points with different belief. We elaborate on three key problems during applying the Dempster-Shafter evidence theory. The approach has been implemented in a practical wireless network and experimental results indicate that it is feasible in both accuracy and locating time. An average location estimation error below 2 meters can be obtained.