Abstract: To distinguish the real modal from the false one and to identify the modal parameters accurately are two key problems in applying ITD as an identification technique. The ITD method uses a confidence factor of the modal to distinguish the real from the false, its accuracy is significantly influnced by the noise in the recorded signal and the numbers of the measured points, and the results have to be judged artificially. This method also requires to change the least square equations into the form of normal equations to solve the matrix of the system, which makes a worse ill-condition and instability of the equations to be solved. In order to improve the ITD method, the real and false modals are distinguished by means of the comparison of the frequence values of different series, with which a quite reasonable reliability can be assured even with less measuring points and the influence of the recorded noise on the real modal of the system can be reduced to a minimum extent. An improved Gram-schmit orthogonalization approach is used to solve the least square equations in the identification calculation for a higher precision. An identification program is designed it with the FORTRAN77 algorithmic language. With the program several tests are performed with satisfactory results.