The classical 2D-ESPRIT algorithm is sensitive to signal-to-noise-ratio (SNR) when estimating the pa-rameters of two dimensional (2D) geometric theory of diffraction (GTD) model． To solve the problem，an im-proved 2D-ESPRIT algorithm，which improves the anti-noise performance and parameter estimation performance ef-fectively，was proposed． Firstly，the improved algorithm obtains a new matrix Y containing the conjugate informa-tion of the original matrix Ee by constructing an exchange matrix． Secondly，a new covariance matrix R can be ob-tained by auto-correlation，cross correlation，superposition and average processing． Finally，the parameters of 2D-GTD model can be estimated by calculating the new covariance matrix R． Simulation results show that the im-proved 2D-ESPＲIT algorithm has a better anti-noise performance，and it owns higher estimation precision and more stable performance than the classical 2D-ESPRIT algorithm accurately in the case of low signal-to-noise ratio． The impact of other factors，such as matrix pencil parameters and paring variable β，on parameter estimation precision
was also studied，which can be provide references for future simulations．