Several recent studies have shown how to properly calculate the observed clustering of galaxies in a relativistic context , and uncovered corrections to the Newtonian calculation that become significant on scales near the horizon . Here , we retrace these calculations and show that , on scales approaching the horizon , the observed galaxy power spectrum depends strongly on which gauge is assumed to relate the intrinsic fluctuations in galaxy density to matter perturbations through a linear bias relation . Starting from simple physical assumptions , we derive a gauge-invariant expression relating galaxy density perturbations to matter density perturbations on large scales , and show that it reduces to a linear bias relation in synchronous-comoving gauge , corroborating an assumption made in several recent papers . We evaluate the resulting observed galaxy power spectrum , and show that it leads to corrections similar to an effective non-Gaussian bias corresponding to a local f _ { NL,eff } \lesssim 0.5 . This number can serve as a guideline as to which surveys need to take into account relativistic effects . We also discuss the scale-dependent bias induced by primordial non-Gaussianity in the relativistic context , which again is simplest in synchronous-comoving gauge .