This paper presents an approach of freeway traffic state estimation based on speed gradient model. Under the sensitivity analysis of model parameters, it is found that error of model estimation is sensitive to the free flow speed and jam propagation speed, which are recommended to be calibrated online. Moreover, the extended Kalman filter and the unscented Kalman filter methods are introduced combined with the speed Gradient model to solve traffic state estimation problems. The real traffic data were used to evaluate the methods. The results indicate that the accuracies of both extended Kalman filter and the unscented Kalman filter are around 85%, while the latter has a slight vantage in estimation accuracy and disadvantage in computing efficiency. The speed gradient model based traffic state estimation method can estimate and track the traffic dynamics effectively, with less model parameters when compared with similar methods.