The Northridge earthquake caused 1.78 g acceleration in the east-west direction at a site in Tarzana, California, located about 6 km south of the mainshock epicenter. The accelerograph was located atop a hill about 15-m high, 500-m long, and 130-m wide, striking about N78??E. During the aftershock sequence, a temporary array of 21 three-component geophones was deployed in six radial lines centered on the accelerograph, with an average sensor spacing of 35 m. Station COO was located about 2 m from the accelerograph. We inverted aftershock spectra to obtain average relative site response at each station as a function of direction of ground motion. We identified a 3.2-Hz resonance that is a transverse oscillation of the hill (a directional topographic effect). The top/base amplification ratio at 3.2 Hz is about 4.5 for horizontal ground motions oriented approximately perpendicular to the long axis of the hill and about 2 for motions parallel to the hill. This resonance is seen most strongly within 50 m of COO. Other resonant frequencies were also observed. A strong lateral variation in attenuation, probably associated with a fault, caused substantially lower motion at frequencies above 6 Hz at the east end of the hill. There may be some additional scattered waves associated with the fault zone and seen at both the base and top of the hill, causing particle motions (not spectral ratios) at the top of the hill to be rotated about 20?? away from the direction transverse to the hill. The resonant frequency, but not the amplitude, of our observed topographic resonance agrees well with theory, even for such a low hill. Comparisons of our observations with theoretical results indicate that the 3D shape of the hill and its internal structure are important factors affecting its response. The strong transverse resonance of the hill does not account for the large east-west mainshock motions. Assuming linear soil response, mainshock east-west motions at the Tarzana accelerograph were amplified by a factor of about 2 or less compared with sites at the base of the hill. Probable variations in surficial shear-wave velocity do not account for the observed differences among mainshock acceleration observed at Tarzana and at two different sites within 2 km of Tarzana.