We present neutrino mass bounds using 900,000 luminous galaxies with photometric redshifts measured from Sloan Digital Sky Survey III Data Release Eight (SDSS DR8). The galaxies have photometric redshifts between z = 0.45 and z = 0.65, and cover 10,000 square degrees and thus probe a volume of 3h^(−3)Gpc^3, enabling tight constraints to be derived on the amount of dark matter in the form of massive neutrinos. A new bound on the sum of neutrino masses of 0.26 eV, at 95% confidence level (CL), is obtained after combining our sample of galaxies, which we call ”CMASS”, with WMAP 7 year Cosmic Microwave Background (CMB) data and the most recent measurement of the Hubble parameter from the Hubble Space Telescope (HST). This constraint is obtained with a conservative multipole range choice of 30 < l < 200 in order to minimize non-linearities, and a free bias parameter in each of the four redshift bins. We study the impact of assuming this linear galaxy bias model using mock catalogs, and find that this model causes a small (1 − 1.5σ) bias in the dark energy density. For this reason, we also quote neutrino bounds based on a conservative galaxy bias model containing additional, shot noise-like free parameters. In this conservative case, the neutrino mass bounds are significantly weakened, e.g. < 0.36 eV (95% confidence level) for WMAP+HST+CMASS (lmax = 200). We also study the dependence of the neutrino bound on multipole range (lmax = 150 vs lmax = 200) and on which combination of data sets is included as a prior. The addition of supernova and/or Baryon Acoustic Oscillation data does not significantly improve the neutrino mass bound once the HST prior is included. A companion paper (Ho et al. 2012) describes the construction of the angular power spectra in detail and derives constraints on a general cosmological model, including the dark energy equation of state w and the spatial curvature ΩK, while a second companion paper Seo et al. (2012) presents a measurement of the scale of baryon acoustic oscillations from the same data set. All three works are based on the catalog by Ross et al. (2011).