Processing of seismic data from LASA can be accomplished by simple delay and sum or beam forming, weighted delay and sum or by maximum‐likelihood filtering. It is found that band‐pass prefiltering the data has a considerable effect on the results. The extra gain produced by maximum likelihood filtering does not justify use of this process on‐line, but could justify off‐line processing in special cases. The spurious precursor introduced by maximum‐likehood processing, and shown to be caused primarily by the signal amplitude scatter within a subarray, can be effectively reduced by using amplitude equalization. The signal amplitude scatter within a subarray, and between subarrays, does not degrade seriously the performance of the signal processing. The effectiveness of the processing depends not only on the array spatial filtering ability, but on differences in absolute level of the noises in the various sensors. Our experience in processing a weak event whose apparent magnitude, as averaged over the LASA was 3.5, indicated that the maximum likelihood filter was capable of extracting the event from the background noise well enough that the event could be identified as an earthquake and not a nuclear explosion.