Abstract This paper considers detection of a signal in underwater colored noise on small aperture array, and an Adaptive Matched Filter based on Maximum Likelihood (ML-AMF) is proposed. The Direction-Of-Arrival (DOA) of signal is firstly estimated, and then energy detection is carried out by using the pre-estimated DOA. The test statistic is deduced. ML-AMF method is robust to the uncertainties of steering vector. Simulation and experiment results show the effectiveness of the method. Experiment results on an 8 element array show that ML-AMF performs better than Minimum Variance Distortionless Response (MVDR) and Conventional BeamForming (CBF) of 1~5 dB and 12~17 dB respectively.