To improve the state estimation accuracy for CIR term structure model of interest rates, this paper establishes the discrete nonlinear filtering formulation of CIR model and then adopts Gaussian particle filter (GPF) to generate the approximate optimal state estimation. Compared with the popular extended Kalman filter (EKF), GPF employs Gaussian distribution based on importance sampling algorithm to approximate the posterior probability while avoiding the error resulting from linearly approximating the function itself. The two nonlinear estimation methodologies are implemented and compared on simulated data. Results are presented to demonstrate the more accurate ability of GPF-based CIR filtering model to describe the dynamics of the term structure of interest rates.