Inconsistent geoid heights may introduce uncertainties that limit its use in applications. Consequently, the scientific community has been collaborating to determine more rigorous geoid models in order to improve the accuracy of orthometric heights estimated with height transformation from the Global Navigation Satellite System (GNSS) positioning techniques. In this regard, this study evaluated the influence of different parameters (such as mass density, topographic models, geopotential models, and Stokes kernel modifications) used in the determination of a gravimetric geoid model through the remove-compute-restore (RCR) technique to contribute to these efforts. As a result of the analyses, it was determined that the Stokes kernel modification type and the global geopotential model selection significantly influence the improvement of geoid models in the study area (state of São Paulo, Brazil). However, the resolution of digital terrain models and the values of the lateral topographic density have less of an impact on the model's accuracy when determining the geoid. In order to acquire findings that are consistent with geoid heights estimated by each determination technique, it is necessary to conduct a rigorous study of the employed data and computation parameters as a preliminary process.