We have defined the complex bed topography for a section of a small temperate glacier using 50 MHz monostatic short-pulse radar data and a synthetic-aperture array-processing method. The data were collected on a 100 m by 340 m array grid in the upper stem of Gulkana Glacier, central Alaska, U.S.A. The array processing was based on a modified three-dimensional (3-D) Kirchhoff migration integral and implemented with a synthetic-aperture approach that uses sequences of overlapping sub-arrays to generate depth images in vertical planes. Typical sub-array beam patterns are generally <5° at the −6 dB level, giving a flashlight-like searching capability without distorting the wavelet shape. The bed topography was constructed using normal reflections picked from 3-D array depth images. In some instances reflections were imaged outside the data-cover-age area. The bed surface dips steeply, both parallel and transverse to the direction of ice flow. The maximum observed depth is roughly 140 m. The 3-D method resolved bed dips up to 45°. In regions of steepest dip, it improved depth accuracy by 36% compared with raw data, and by 15% compared with standard two-dimensional (2-D) migration. Over 12 dB of signal-to-noise improvement and improved spatial resolution was achieved compared to raw data and 2-D migration. False bottom layering seen in the raw data and in 2-D migrations is not observed in the 3-D array results. Furthermore, loss of bottom reflections is shown by the 3-D migration to be attributable to the dip and curvature of the reflector, and not scattering losses or signal clutter from englacial inclusions.