Determining the location of nodes with error as low as possible in wireless sensor networks is an essential task for many applications. It can be shown that, in some case, evident location estimation errors would be brought even if the ranging error is low. This work involves a parameter called geometry favorable factor (GFF) to filter these ill case. We propose an improving method choosing the available beacon nodes (nodes with prior known locations) with the best GFF for sensor location estimation. Simulation results show that the improving method lead to a significant improvement in location estimation accuracy compared with normal method that without filtering, while some decrease in population of localizable nodes.