The observability and measurement accuracy are low in single observer passive location, so the initial error is usually large. As the unscented kalman filter(ukf) in single observer passive location is sensitive to the initial value and will divergent because of the numerical calculation error, an improved forward-backward smoothing algorithm based on square-root unscented kalman filter(srukf) is presented. To guarantee the stability of filter, the algorithm used the covariance square root matrix instead of covariance matrix in the process of estimation. And the algorithm utilized backward smoothing to get a more accurate state estimate as an initial condition to improve the robustness of the initial value. Simulation results show that the algorithm has better performance to ukf and srukf in the filter’s stability, convergence velocity, positioning precision and the robustness of the initial value.