Abstract Some laboratory models of slip find that a critical amount (or velocity) of slow slip is required over a nucleation patch before dynamic failure begins. Typically, such patch sizes, when extrapolated to earthquakes, have been thought to be very small and the precursory slip undetectable. Ohnaka (1992, 1993) has proposed a model in which foreshocks delineate a growing zone of quasi-static slip that nucleates the dynamic rupture and suggests that it could be large enough (∼10 km across) to be detectable and thus useful for short-term earthquake prediction. The 1992 Landers earthquake (M 7.3) had a distinctive foreshock sequence and initiated only 70 km from the strain meters at the Piñon Flat Observatory (PFO). We use this earthquake to investigate the validity and usefulness of Ohnaka's model. The accurate relocations of Dodge et al. (1995) show that the foreshock zone can be interpreted as expanding from an area of 800 m (along strike) by 900 m (in depth), to 2000 by 3200 m in the 6.5 hr before the mainshock. We have calculated the deformation signals expected both at PFO and 20 km from the foreshock zone, assuming either constant slip or constant stress drop on a circular patch expanding at 5 cm/sec over 6.5 hr. We find the slips or stress drops would have to have been implausibly high (meters or kilobars) to have been detectable on the strain meters at PFO. Slightly better limits are possible only 20 km from the source. Even though the distance from Landers to PFO is small compared with the average spacing of strain meters in California, we are unable to prove or disprove Ohnaka's model of earthquake nucleation. This suggests that even if the model is valid, it will not be useful for short-term prediction.