Combining measurements of the galaxy power spectrum and the cosmic microwave background (CMB) is a powerful means of constraining the summed mass of neutrino species , but is subject to systematic uncertainties due to non-linear structure formation, redshift-space distortions and galaxy bias. We empirically test the robustness of neutrino mass results to these effects by separately analysing power spectra of red and blue galaxies from the Sloan Digital Sky Survey (SDSS-II) Data Release 7 (DR7), combined with the CMB 5-yr Wilkinson Microwave Anisotropy Probe (WMAP5) data. We consider fitting for a range of maximum wavenumber k using 12 different galaxy bias models. For example, using a new model based on perturbation theory and including redshift-space distortions, the all-galaxy power spectrum combined with WMAP5 for a wavenumber range of k < 0.2 h Mpc−1 yields 95 per cent confidence limit eV. The red and blue galaxy power spectra give 0.41 and 0.63 eV, respectively, for this model. Using mock catalogues, we find the expected difference in these limits assuming a true neutrino mass of zero is 0.10 ± 0.14 eV. Thus, the difference of 0.22 eV between upper limits on neutrino mass for red and blue galaxies is approximately 1σ from the expected value. We find similar results for the other models and k ranges tested. This indicates good agreement for current data but hints at possible issues for next-generation surveys. Being able to perform such systematic tests is advantageous, and future surveys would benefit by including broad galaxy populations and luminosities that enable such a decomposition.