Analysis of refracted first breaks has traditionally assumed that the data have been collected using special geometries that enhance the results, e.g. reversed profiles with regularly spaced geophones in-line between two shots. The geometries used to collect most 3D reflection data are quite different. Thus refraction analysis to obtain statics using traditional methods requires both approximations and selection of a subset of the data. The surface-consistent method allows use of nearly all the data, thus providing the high redundancy required for statistical robustness. A large survey incorporating both a new 3D survey and older 2D lines yielded refraction statics that greatly improved the final results.The surface-consistent method assumes that the refractor can be approximated by a horizontal plane under each station. When the refracting surface is steeply dipping, this assumption may break down. The Generalized Reciprocal Method is a traditional refraction analysis technique that gives improved results for steeply-dipping refracting surfaces. The surface-consistent method can also be made less sensitive to dip by a generalization, i.e. assume a dipping plane under each station. The refracting surface between the stations can be approximated by interpolation of these planes using the "linear projection" technique. A comparison of results using the two surface-consistent methods on synthetic data generated from a model of a buried, steep-walled valley shows the superior results obtainable.