Laboratory and theoretical studies suggest that there is a minimum patch size for rupture on faults, thus limiting observed earthquake sizes to above some moment or magnitude corresponding to this patch dimension. This limit, most likely being at a much lower magnitude than that which can be observed due to background noise of the Earth and to attenuation, is difficult to verify for real earthquakes. We have recorded earthquakes to as small as M -2 (Richter local magnitude) in the aftershock zone of the M 5.6 Little Skull Mountain earthquake of 29 June 1992. The network threshold for location is below M 0. However, by considering all the triggered earthquakes at station LSC, which is just above the aftershock zone, and combining these with the located events, the recurrence curve is extended well below the network threshold. We do this by forming a relation between the trace amplitudes of the LSC recordings and the network magnitudes for larger events and then assigning an estimated M to the smaller triggered events. The recurrence curve plotted for the combined data shows a constant b-value of 0.82 down to roughly M -1.2. We show that this magnitude is somewhat below the detection threshold of LSC. The smallest recorded events are similar in appearance to events larger by as much as 3 magnitude units in the aftershock zone, partly due to the band limitation of the 100-samples/sec data and local attenuation. S-to-P ratios much greater than 1 and broadband signals for these small events are indicative of normal tectonic earthquakes. The seismic moments of the very smallest events near M -2 are approximately 5 × 109 N m, larger than observations in deep mines by at least an order of magnitude. Because the source corner frequency is too high to be seen in the spectrum of the smallest earthquakes, we cannot estimate source radii.