Channel mismatch may severely degrade the performance of GNSS antenna arrays, thus it must be calibrated in GNSS applications. Estimation of channel mismatch is the precondition and key of channel calibration. The estimation precision will influence the effect of calibration directly. However, few works on mismatch estimation could be found in previous literatures. In this paper, based on the modeling of the array channels, two kinds of method are proposed to estimate channel mismatch. Theoretical and simulative analysis show the feasibility and the good performance of the proposed methods which can make the estimation precision achieves the Cramer-Rao Bound (CRLB) asymptotically. Given input Signal-to-Noise ratio (SNR) equal to 20 dB and the data length of 1024, the variance of mismatch estimation can be reduced to the level of 10−4 according to numerical analysis and practical demonstration. The technique presented in this paper can be adopted to the calibration of antenna arrays.