An accurate modeling of geological and geometallurgical variables is essential for reliable mineral resource evaluation, mining project development and mining operations. Gaussian transformations, like the univariate normal score transformation (NST) and the projection pursuit multivariate transformation (PPMT), help to prepare data for geostatistical simulations. This study compares NST and PPMT through two real case studies: a stratabound copper deposit and a nickel laterite deposit. While PPMT reproduces collocated statistical relationships much better than NST, it turns out to alter spatial structures. We furthermore explore two practical solutions to enhance NST: use of service variables and geological domaining. The former helps to reproduce physical constraints, while the latter ensures that different data populations are modeled separately. Our findings suggest that no single Gaussian transformation method works best in all cases. Instead, problem-specific solutions can significantly enhance the realism of geostatistical models by better reproducing both statistical and spatial dependence relationships.