The effect of local strucuture on the characteristics of the regional phases Pg and Lg was investigated by analyzing events at epicentral distances less than 10 ¿ from the Nevada Test Site. The stations used are on markedly different geological structures; three on granitic intrusions, two on Pahute Mesa, a feature overlaying a buried volcanic caldera, and, finally, four on Yucca Flats, a deep alluvial valley. The crustal effect on amplitudes is quite significant for both Lg and Pg; Yucca Flats gives amplitudes as much as a full order of magnitude higher than the stations on granite, while Pahute Mesa stations are 0.7--0.8 magnitude units higher in amplitude than the granite sites. Besides gross amplitude effects, the local crust also induces prolonged ringing at Yucca Flats, resulting in the lengthening of the signal wave trains. The observed site amplification effects show no clear dependence on the azimuths of arrivals, indicating that the phenomenon is due to local structures and is not caused by lateral refraction of surface wave modes. The observed amplification is similar in magnitude for the vertical component of both Pg and Lg, thus not affecting the Pg/Lg amplitude ratio, a possible discriminant between earthquakes and explosions at regional distances. Finite difference theoretical simulations using two-dimensional models of the Yucca basin reproduce, in a qualitative sense, the ringing effect observed at the Yucca Flats. Any attempt at matching characteristics of Pg and Lg with synthetic seismograms computed for laterally homogeneous layered structures is apt to lead to false conclusions about the source and path properties unless the recording sites used are carefully chosen for small crustal effects.