While stellar rotation periods P_rot_ may be measured from broad-band photometry, the photometric modulation becomes harder to detect for slower rotators, which could bias measurements of the long-period tail of the Prot distribution. Alternatively, the P_rot_ distribution of stars can be inferred from their projected rotation velocities vsin i and radii R, without being biased against photometrically quiet stars. We solve this inference problem using a hierarchical Bayesian framework, which (i) is applicable to heteroscedastic measurements of vsin(i) and R with non-Gaussian uncertainties and (ii) does not require a simple parametric form for the true P_rot_ distribution. We test the method on simulated data sets and show that the true P_rot_ distribution can be recovered from ~> 100 sets of vsin i and R measured with precisions of 1 km.s^-1^ and 4 per cent, respectively, unless the true distribution includes sharp discontinuities. We apply the method to a sample of 144 late-F/early-G dwarfs in the Kepler field with vsin(i) measured from Keck/HIRES spectra, and find that the typical rotation periods of these stars are similar to the photometric periods measured from Kepler light curves: we do not find a large population of slow rotators that are missed in the photometric sample, although we find evidence that the photometric sample is biased for young, rapidly rotating stars. Our results also agree with asteroseismic measurements of P_rot_ for Kepler stars with similar ages and effective temperatures, and show that =~ 1.1 M_{sun}_ stars beyond the middle of their main-sequence lifetimes rotate faster than predicted by standard magnetic braking laws.