Short-tem traffic flow data is characterized by high volatility and noninearity. it reflects the nature of frequent congestion in the lane that shows a strong nonlinear feature. traffic state estimation based on the data obtained by electronic sensors is critical for many intelligent traffic management and traffic control system. in this paper, a solution to freeway traffic estimation is proposed using an impoved version of particle filter (PF) and a macrooscopic traffic models as well as non-gaussian signals. experiments are conducted based on the data obatined from a beijing freeway to evaluate the robustness and generality of the proposed method. the experimental result show that the proposed technique can produce accurate estimation.