In the context of measuring structure and morphology of intermediate redshift galaxies with recent HST/ACS surveys, we tune, test, and compare two widely used fitting codes (GALFIT and GIM2D) for fitting single-component Sersic models to the light profiles of both simulated and real galaxy data. We find that fitting accuracy depends sensitively on galaxy profile shape. Exponential disks are well fit with Sersic models and have small measurement errors, whereas fits to de Vaucouleurs profiles show larger uncertainties owing to the large amount of light at large radii. We find that both codes provide reliable fits and little systematic error, when the effective surface brightness is above that of the sky. Moreover, both codes return errors that significantly underestimate the true fitting uncertainties, which are best estimated with simulations. We find that GIM2D suffers significant systematic errors for spheroids with close companions owing to the difficulty of effectively masking out neighboring galaxy light; there appears to be no work around to this important systematic in GIM2D's current implementation. While this crowding error affects only a small fraction of galaxies in GEMS, it must be accounted for in the analysis of deeper cosmological images or of more crowded fields with GIM2D. In contrast, GALFIT results are robust to the presence of neighbors because it can simultaneously fit the profiles of multiple companions thereby deblending their effect on the fit to the galaxy of interest. We find GALFIT's robustness to nearby companions and factor of >~20 faster runtime speed are important advantages over GIM2D for analyzing large HST/ACS datasets. Finally we include our final catalog of fit results for all 41,495 objects detected in GEMS.