We determine 37 differential extinctions in 23 gravitational lens galaxies over the range 0 < z_l < 1. Only 7 of the 23 systems have spectral differences consistent with no differential extinction. The median differential extinction for the optically-selected (radio-selected) subsample is E(B-V)=0.04 (0.06) mag. The extinction is patchy and shows no correlation with impact parameter. The median total extinction of the bluest images is E(B-V)=0.08 mag, although the total extinction distribution is dominated by the uncertainties in the intrinsic colors of quasars. The directly measured extinction distributions are consistent with the mean extinction estimated by comparing the statistics of quasar and radio lens surveys, thereby confirming the need for extinction corrections when using the statistics of lensed quasars to estimate the cosmological model. A disjoint subsample of two face-on, radio-selected spiral lenses shows both high differential and total extinctions, but standard dust-to-gas ratios combined with the observed molecular gas column densities overpredict the amount of extinction by factors of 2-5. For several systems we can estimate the extinction law, ranging from R_V=1.5+/-0.2 for a z_l=0.96 elliptical, to R_V=7.2+/-0.1 for a z_l=0.68 spiral. For the four radio lenses where we can construct non-parametric extinction curves we find no evidence for gray dust over the IR-UV wavelength range. The dust can be used to estimate lens redshifts with reasonable accuracy, although we sometimes find two degenerate redshift solutions.