Abstract: The actual steering vector with errors is usually modeled as an uncertainty set in most robust adaptive beamforming (RAB) against the steering vector mismatch. However, it is commonly difficult to determine the constraint parameter in practice. In this paper, a RAB based on modified noncircular complex fast independent component analysis (MNC-FastICA) algorithm is proposed. The weight vector of beamformer can be constructed with the separation matrix found by MNC-FastICA algorithm and the amplitude and phase ambiguities of estimations resulted from separation are also calibrated. Thus, the signal directions of arrival (DOA) do not need to be predestinated, which voids the mismatch of signal steering vector due to the error of DOA in classical RAB methods. Moreover, the proposed method is not sensitive to the amplitude and phase errors of array channel so that array calibration is not necessary. Simulations are run and the performances are compared with classical methods such as worst-case performance optimization (WCPO). Results demonstrate the effectiveness and robustness of our method.