Detection methods in earthquake seismology have been dominated in the past by simple energy detectors, often combined with array processing (beamforming) to detect at lower thresholds. Increasingly, correlation detectors now are being employed to exploit the additional gain possible with coherent detection methods. Coherent detection of transient seismic signals is complicated by the fact that earthquake signals generally cannot be predicted with certainty in advance, except for those cases where observations of previous events are available. One approach to solving this problem is to employ subspace detectors that allow variation in the correlation template to match the range of signals anticipated from a specific source region. Subspace templates have been produced empirically from past observations of repeating sources. The current challenge is to compute the templates in areas without past event observations. One approach to doing so calculates the target signal space by forward simulation through many earth realizations drawn from a stochastic earth model. In this talk, we present the results of empirical subspace detection with arrays of sensors (combining array gain with coherent matched-signal gain) and discuss preliminary results from the use of computed subspaces. [This work was performed under the auspices of the U.S. Department of Energy by the University of California, Lawrence Livermore National Laboratory under Contract No. W-7405-Eng-48.]