In this paper, a new packet loss probability (PLP) estimation method is developed based on the Kalman estimation procedure. The Kalman filter is used for the optimal and recursive estimation of the network traffic mean and standard deviation. The estimation is obtained from past measurements to calculate the one-step prediction and is then applied to calculate the PLP parameter with the large-buffer-based estimation formula based on the large deviation theory. The algorithm recursively runs and is applied for the online packet loss control loop, which keeps the PLP parameter under the limit negotiated in the service level agreement with the customer by the Internet service provider. A series of experiments were conducted to evaluate its performance on the live NCIT*net2 network under different traffic arrival models, for two different aggregated traffic rates and different buffer sizes. The numeric results presented in this paper demonstrate the accuracy and effectiveness of the algorithms introduced.