Careful analysis of strong-motion recordings of 13 medium magnitude earthquakes (3.7 ≤ M ≤ 6.5) in the Parkfield, California, area shows that very modest levels of shaking (approximately 3.5% of the acceleration of gravity) can produce observable changes in site response. Specifically, I observe a drop and subsequent recovery of the resonant frequency at sites that are part of the USGS Parkfield dense seismograph array (UPSAR) and Turkey Flat array. While further work is necessary to fully eliminate other models, given that these frequency shifts correlate with the strength of shaking at the Turkey Flat array and only appear for the strongest shaking levels at UPSAR, the most plausible explanation for them is that they are a result of nonlinear site response. Assuming this to be true, the observation of nonlinear site response in small (M <5) earthquakes implies that nonlinear site response can occur at much lower levels of shaking than previously believed. Below I present observations of a resonant frequency shift during five M ≤ 5 earthquakes near Parkfield, California, strongly contrasting with previous studies that have only identified nonlinear site effects for much larger events. In addition to the nonlinear effects seen for the smaller events, nonlinear site response is also observed for two largest earthquakes in the region during the study period (the 2003 M 6.5 San Simeon earthquake and the 2004 M 6 Parkfield earthquake).