Frequency-dependent crustal attenuation (1/Q) is determined for seven distinct physiographic/tectonic regions of the continental United States using high-quality Lg waveforms recorded on broadband stations in the frequency band 0.5 to 16 Hz. Lg attenuation is determined from time-domain amplitude measurements in one-octave frequency bands centered on the frequencies 0.75, 1.0, 3.0, 6.0, and 12.0 Hz. Modeling errors are determined using a delete-j jackknife resampling technique. The frequency-dependent quality factor is modeled in the form of Q = Q₀ fη. Regions were initially selected based on tectonic provinces but were eventually limited and adjusted to maximize ray path coverage in each area. Earthquake data was recorded on several different networks and constrained to events occurring within the crust (<40 km depth) and at least m_b 3.5 in size. A singular value decomposition inversion technique was applied to the data to simultaneously solve for source and receiver terms along with Q for each region at specific frequencies. The lowest crustal Q was observed in northern and southern California where Q is described by the functions Q = 152(±37)f^0.72(±0.16) and Q = 105(±26)f^0.67(±0.16), respectively. The Basin and Range Province, Pacific Northwest, and Rocky Mountain states also display lower Q and a strong frequency dependence characterized by the functions Q = 200(±40)f^0.68(±0.12), Q = 152(±49)f^0.76(±0.18), and Q = 166(±37)f^0.61(±0.14), respectively. In contrast, in the central and northeast United States Q functions are Q = 640(±225)f^{0.344(±0.22)} and Q = 650(±143)f^0.36(±0.14), respectively, show a high crustal Q and a weaker frequency dependence. These results improve upon previous Lg modeling by subdividing the United States into smaller, distinct tectonic regions and using significantly more data that provide improved constraints on frequency-dependent attenuation and errors. A detailed attenuation map of the continental United States can provide significant input into hazard map mitigation. Both scattering and intrinsic attenuation mechanisms are likely to play a comparable role in the frequency range considered in the study.