We validate a new method to determine surface-wave attenuation from seismic ambient noise, both numerically and by application to recordings from a dense broadband array. We generate synthetic recordings of numerically simulated ambient seismic noise in several experimental setups, characterized by different source distributions and different values of attenuation coefficient. We use them to verify that: (I) “cross-terms” cancel out, as predicted by the theory; (II) the source spectrum can be reconstructed from ambient recordings, provided that the density of sources and the attenuation coefficient are known; (III) true attenuation can be retrieved from normalized cross correlations of synthetic signals. We then apply the so validated method to real continuous recordings from 33 broadband receivers distributed within the Colorado Plateau and Great Basin. A preliminary analysis of the signal-to-noise ratio as a function of azimuth reveals a SW-NE preferential directionality of the noise sources within the secondary microseism band (6-29 8 s), as previously reported by other authors. By nonlinear inversion of noise data we find the attenuation coefficient in the area of interest to range from ∼ 1 × 10 −5 m−1 at 0.3 Hz to ∼ 4.5 × 10 −7 m−1 at 0.065 Hz, and confirm the statistical robustness of this estimate by means of a bootstrap analysis. This result is compatible with previous observations made on the basis of both earthquake-generated and ambient Rayleigh waves. In this regard, the new method proves to be promising in accurately quantifying surface wave attenuation at relatively high frequencies.