The paper deals with geoid modeling using the least-squares modification of the Stokes' and Hotine's formulae using gridded and non-gridded terrestrial gravity data over the Auvergne test area (France). Differently from the conventional way of using gridded gravity data, this study utilizes non-gridded data as the input to the geoid modeling and compare the differences between the computational approaches in terms of the geoid model accuracies. For this purpose, the calculated gravimetric geoid models are validated at high-accuracy GNSS/leveling benchmarks provided in the test dataset. The geoid models computed using the grid gravity data yielded 3.8 cm and 4.0 cm accuracies before fitting for Hotine's and Stokes' methods, whilst the evaluation of the geoid models with non-gridded data yielded 6.6 cm and 5.7 cm, respectively. The numerical comparisons of the different geoid models are presented and discussed in detail.