Horse Mesa Dam is located in the Mazatzal Mountains about 60 miles east of Phoenix, Arizona and is a part of the Salt River Project. Early in 1981, the management of the Salt River Project decided to solve problems with the existing roadway to the dam by building an improved roadway and retaining wall. Geotechnical engineers on the project originally planned to obtain depth-to-bedrock information for the design of these structures by drilling a large number of boreholes to bedrock. However, we decided that more complete information could be obtained less expensively by using fewer drill holes tied together with seismic refraction lines. The complicated physical setting of the site made planning for the survey very important. The bedrock surface at the site is quite irregular, dipping as steeply as 30 degrees across the axis of the roadway, and the space between the stream bank on one side of the roadway and the canyon wall on the other side is very narrow. Because of these constraints, we calculated that head waves refracted along the bedrock surface would never be observed on refraction lines shot across the roadway. Therefore, we decided to shoot two long refraction lines along the axis of the existing roadway--one each on the inside and outside shoulders of the road bed--to provide two points on the bedrock surface at any particular cross-section of the road. Field work at the site was done between June 10 and June 15, 1981. We obtained reversed coverage on 914 m (3 000 ft) of 70 m (230 ft) spreads along the road. Because of the attenuation of the coarse overburden material, 0.45 kg charges of Kinestik were necessary to deliver seismic energy across these spreads, and 0.15 kg charges were necessary at the center shots. To avoid overwhelming background noise at the site, we performed much of the survey in the evening or early morning when the turbines at the dam were idle.Survey results showed not only a highly irregular bedrock surface, but also indicated substantial velocity variations within both the overburden and bedrock. Overburden velocities varied from 366 m/s (1 200 ft/s) to 549 m/s (1 800 ft/s), while bedrock velocities varied from 1 829 m/s (6 000 ft/s) to nearly 7 620 m/s (25 000 ft/s). We interpreted the results with the generalized reciprocal method (GRM), which is widely used in Australia but is not so well-known in the United States. This method proved to be particularly well-suited for situations involving large lateral velocity variations and highly irregular refractor surfaces. When compared with the seven borings that were eventually done on the roadway, the refraction interpretation was accurate to within a few feet; or, in terms of the nominal depth to bedrock, the results were accurate to within about 10 percent. This is remarkable agreement considering the irregularity of the bedrock surface. The Horse Mesa Dam study shows the importance of planning a survey with the physical characteristics of the site and the goals of the survey in mind. It also shows the value of a powerful, flexible interpretation method--in this case, the GRM. Most important, it demonstrates that engineering geophysics is capable of providing useful, accurate information at a lower cost than drilling alone.