This paper presents a robust distributed weighted least-square scaling localization algorithm based on the local network feature. It first filters out 1-hop gross ranging errors of the unknown nodes by using the threshold parameters of extreme ranging overestimates and extreme ranging underestimate with the local network feature, and then employs Gauss-kernel- weighted least squares to position nodes. Simulation results confirm that this localization scheme outperforms traditional weighted least squares, which does not adopt outlier identification scheme. The localization accuracy of the unknown nodes which are directly related to the outliers improves remarkably.