To solve the problem of beam selection or capturing the highest possible signal power, we propose a sequential test that can adapt to the SNR operating point and speed up the selection procedure in terms of the number of required observations in comparison to a perfectly tuned fixed length test assuming genie knowledge. The speed up gets higher for lower SNR and becomes of particular interest in massive Multiple Input Multiple Output (MIMO) systems using beamforming, where the number of candidate beams is large and exhaustive search can cause intolerable delay due to limited channel coherence time