Abstract: To solve the problems of high model order of acoustic relative transfer function and low signal to noise ratio of identification data, a highly accurate parametric identification method of acoustic relative transfer functions is proposed by means of frequency domain system identification. Firstly, an errors-in-variables identification framework is established, and by taking a periodic chirp signal as sound source excitation, the maximum likelihood formulation is given to estimate acoustic relative transfer functions. Then, the orthogonal Forsythe polynomial is used to solve the numerical problem of excessive number of Jacobian matrix conditions caused by high-order systems, and a strategy for generating the initial values required by the maximum likelihood method is provided. Finally, the effectiveness of the proposed method in identifying the acoustic relative transfer function under reverberant environment is verified by simulation example and experimental tests.