Geostatistically simulated representations of mineral deposits are used for resource/reserve assessments, mine design and production planning, project evaluation and so on. Categorical simulation methods are part of the related framework and are used to generate realizations of geological units and their boundaries. The simulation of the geological units of a gold deposit is presented in this paper, based on a data-driven, high-order geostatistical simulation method. The latter method makes neither distributional assumptions nor utilizes a training image, but consistently uses high-order spatial statistics from the available data. The application at the Saramacca gold deposit, Suriname, shows the practical aspects of the method, including the validation of simulated realizations of geological units through the reproduction of the high-order statistics of the data.