To deal with the problems of complex structure parameters, unknown interference factors and lower modeling precision in the Dynamically Tuned Gyroscope (DTG) mechanism modeling by the traditional Subspace Identification Method(SIM), an improved SIM was proposed. Firstly, the DTG state space model sets were determined and DTG inherent colored noises were discussed. Then, the traditional SIM were modified for the colored noise problem and the orthogonal projection of a data Hankel matrix was used to eliminate the SIM bias of traditional method. Finally, a confidence ellipse was introduced in the numerical simulation to analyze the statistics feature of the modified algorithm. Simulation results indicate that the identified results of modified algorithm are unbiased at different colored noise influences, and the variance is related to the noise strength and data length. The identification experiments show that the identification performance of modified SIM is apparently better than that of the traditional SIM, and the identification fitting degree is more than 90%, which means that the modified algorithm is suitable for the DTG system modeling.