Abstract: The ambient noise is the coherent noise in the actual shallow water waveguide, and the minimum variance matched field source power estimation method can accurately estimate the radiated power of the acoustic source under the background of coherent noise, but this method is greatly affected by the environmental uncertainty. In addition, the signal amplitude is used as an intermediate quantity to estimate the source power, thus the signal amplitude estimation error will be transmitted to the result of the source power estimation. In this paper, a covariance matrix fitting robust minimum variance matched field source power estimation method is proposed. The uncertain set of the acoustic transfer function is introduced in this method, and by using the covariance matrix fitting, the sound source power estimation problem is modeled as the problem of taking the extreme value of the function under the constraint of the uncertain set of the acoustic transfer function. This problem can be solved by Lagrange multiplier method to obtain the estimate of the acoustic transfer function and the source power. The fundamental reason that the environmental mismatch affects the performance of radiated power estimation of the source is the large deviation of the acoustic transfer function, and the deviation of the acoustic transfer function is effectively reduced, which improves the environmental mismatch robustness significantly. In addition, this method can directly estimate the source power without using the signal amplitude as an intermediate quantity, thus avoiding the transmission of estimation error. Simulation results demonstrate the robustness of the covariance matrix fitting robust minimum variance matched field source power estimation method.