A linear adaptive algorithm was developed for array beamforming purposes. The design goal for the algorithm is to minimize the squared filter output subject to filter constraints which allow energy propagating from the array steering direction to pass without being distorted. The adaptive filter coefficients were initialized to satisfy the constraints which were preserved during the iterations. The adaptation rate is inversely varied with filter output and total input channel power. Performance of the algorithm was studied using the recorded short-period array data from the Korean Seismic Research Station. Processed were a high-amplitude signal from Kamchatka, a medium-amplitude signal from eastern Kazakh, and a number of low-amplitude signals from central Eurasia. Results of signal-to-noise ratio gain relative to a conventional beamformer among the events tested were consistent and were in the range of 4.5 to 6.5 dB in the wide passband. Much better signal-to-noise ratio improvement was obtained in the low-frequency passband. The adaptive algorithm was programmed in the real-time mode and can be implemented in a front-end detection system.