Receiver-function methods utilize three-component teleseismic P coda to determine structure beneath seismographs. Most methods treat coda complexity as mode conversions and reflections off subhorizontal interfaces beneath the station, such as the Moho. This assumption is evaluated beneath a broadband array in the Kopet Dagh foothills, at Geyokcha, Turkmenistan, by applying array processing techniques to P coda waves at frequencies near 1 Hz. Beneath the Geyokcha array, horizontal components show signals that deviate significantly in slowness and bearing from predictions of a plane-layered Earth. Large phases arrive at the same time as the predicted Ps conversion from the Moho yet systematically come from a single azimuth (from NNE) and travel at phase velocities of 3 to 5 km/sec. The coda waves appear to be P-to-S converted waves from a near-surface scatterer. An inversion for scatterer location is developed and applied to 29 events. Most coda energy comes from a scatterer ∼16 km NNE of the array. This location is coincident with the range-bounding faults that separate Kopet Dagh mountains from adjacent basins. Thus, array analysis shows that signals potentially interpreted as Moho conversions are instead scattered from a near-surface structure. Although these results suggest caution is necessary in interpreting single-station receiver functions, particularly at high frequencies, they also demonstrate the power of array sampling in resolving lateral structure variations.