The paper describes a new technique for adaptive antenna beamforming. By analysing the adaptive antenna as a least-squares problem, it is possible to organise the control processor around a recursive minimisation procedure operating in the data domain. This avoids the explicit computation of the covariance matrix estimate as required by the method known as sample matrix inversion (SMI). The paper further shows how the least-squares processing algorithm may be implemented in an efficient pipeline architecture using a triangular systolic array. Computer simulation results are presented which show the performance of this new technique compared with sample matrix inversion.