Abstract: A multi-subarray subspace fitting method, which considers mutual coupling among array elements, to estimate the directions of arrival is presented. The mutual coupling matrix of uniform linear arrays is modeled with a banded symmetric Toeplitz matrix. According to the freedom of mutual coupling matrix, part of the array elements at the two sides is neglected. The theoretical characteristics of ESPRIT algorithm, one of the subspace.fitting algorithm, are studied. For large N and uniform linear array, the MCLS-ESPRIT estimation error is asymptotically jointly Gaussian distributed with zero means, and its covariances is obtained. It is known from the simulation that there exists a subarray with lowest estimation error for certain number of the array elements and the signal sources, and its performance of estimating the directions of arrival is close to the ideal situation when mutual coupling does not exist while its variance has some increase. The method presented in this paper has been applied in high resolution bathymetry sidescan sonar and good results are obtained. At the cost of increasing the number of the array elements, the method presented in this paper can reduce the affection of the mutual coupling among array elements.