A closed-loop recursive subspace identification algorithm is proposed to identify the closed-loop systems online by using orthogonal decomposition method. Firstly, the deterministic-stochastic model is built according to the projection relationship among the input, output and exogenous signals, and the recursive QR decomposition is achieved through GIVENS reduction. Then the recursive subspace identification with forgetting factor is used to estimate the extended observability matrix and reduce the computation of QR and SVD decomposition. Finally, the identification experiments are conducted on a gyro closed-loop system. The experiment results show that the identification fitting degree is more than 91% and this algorithm can be used to identify the gyro system model parameters.