Surface and subsurface high-frequency (>1 Hz) noise data were recorded using nearly identical instrumentation at three widely separated sites in the United States (Amarillo, Texas; Datil, New Mexico; and Pinedale, Wyoming) for extended periods of time under varying wind conditions. While the sites are geologically distinct, the near-surface noise environments have many common features that we believe may be due in large part to the presence of a surficial layer of highly attenuative unconsolidated material at each site. Noise levels seen at or near the surface (5 m or less) are much higher (up to 30 dB) and much more variable (power range up to 44 dB) than those seen at depth (the smallest range was 9 dB for 1951 m at Amarillo). The greatest gains in noise level reduction are realized within the first 100 m and probably much shallower (< ~10 m). Regardless of the wind profile or local lithology, all sites show an excellent correlation between increased noise levels and higher wind speed, even at significant depths (367 m at Amarillo). Wind-generated noise is broadband (at least 15 to 60 Hz) and apparently nonlinear, increasing dramatically when a wind speed threshold is exceeded (3 to 4 m/sec within a few meters of the surface; as high as 8 m/sec at a depth of a few hundred meters). It is possible to be essentially completely shielded from the wind-generated component of seismic noise by deploying instruments at sufficient depth, but we observed this only for the two deepest deployments (1219 and 1951 m, both at Amarillo). Reducing the wind profile at the surface, however, can yield similar reductions for a much smaller cost. Cultural or ``workday'' noise, if present (depending on the remoteness of the site), is typically much weaker (10 dB or less) than wind noise but may propagate very effectively to great depths and therefore could be of concern for very deep deployments where wind is not a factor.