The precision with which the elevation of a feature, such as a terrace, can be measured depends on the characterization of the noise contaminating the measurement. A method for identification and extraction of terrace feature elevations is presented and the topographic noise, due to erosion, as well as measurement error, is quantified. High-resolution digital elevation models (DEM) are acquired at six wave-cut, volcanic bedrock terrace sites from around the highstand of paleo-Lake Lahontan in the western Great Basin. Local DEMs, which are tied to regional geodetic control, were acquired using conventional total station, rapid postprocessed and real-time kinematic Global Positioning System methods. The topographic signal is processed with derivative filters for geomorphic feature recognition and averaging for noise reduction. Results indicate that noise levels for identifiable features such as riser crest, knickpoint, and slope inflection point are statistically equivalent and on the order of 0.5 m standard deviation. Averaging within topographic bins spanning ~50 m along terrace strike yields feature elevation estimates with standard errors on the order of 0.12 m. The mean bench window elevation (between the riser crest and knickpoint) has the lowest standard error and is systematically related to water level. Propagation of surveying, geoid estimation, and terrace feature elevation estimation errors indicates that displacements on the order of 0.5 m may be resolved using these methods. Elevation estimate interpretation involves terrace development, degradation, and neotectonics, but this new methodology has significant advantages in studies of neotectonic or geomorphic processes using local terrace elevation measurements. ¿ 2001 American Geophysical Union