Empirical Matched Field Processing (EMFP) is a narrow frequency band method for signal detection and estimation which recognizes a characteristic phase and amplitude structure over an array of sensors. EMFP can provide an enhanced capability for detecting and identifying signals of special interest from a target source region and we here explore the extent to which it can constitute a comprehensive front-end detector for general seismic signals. On small aperture seismic arrays, we demonstrate that a relatively small number of matched field templates are required to cover the parameter space of anticipated signals. In addition, estimates of apparent velocity and back azimuth made directly from the empirical steering vectors are likely to be far more accurate than those made from classical array processing since the bias resulting from path and near-receiver characteristics are corrected for empirically. On larger aperture arrays, significantly more templates are required to provide the same coverage of the anticipated parameter space. On seismic arrays for which signal coherence in the frequency band of interest is poor, a given matched field detector will provide a very signal-specific detector and vast numbers of templates would be required for a comprehensive detector. We demonstrate how classical slowness estimates for high frequency signals of interest can be enhanced by considering only selected sensor pairs and recommend similar approaches for mitigating signal incoherence for general matched field detectors.