The golden-faced saki monkey Pithecia pithecia chrysocephala (Cebidae, Primates) was observed eating soil from termite nests during a long-term study of a family group in a Central Amazonian forest fragment. In this paper we describe the behaviour involved in the geophagy in these monkeys, and the results of geochemical and physical analyses of the termite nest material, as well as root mat and topsoil samples below the trees, in order to clarify the possible reasons for it. The sakis ate soil from nine arboreal termite nests on 26 soil feeding-bouts (in 853 observation hours); 25 soil feeding-bouts occurred in March 1987 (rainy season), during 19 days or 132 observation hours, and occupied 0.7% of the feeding time. Geophagy frequencies did not differ between sexes (17 feeding-bouts of four females and 8 for two males). Mineral composition was higher in arboreal termitaria than in the topsoil. Kaolinite was the major clay component. Tannin adsorptive capacity, tested through a modified radial diffusion method of Hagerman, was around 10–20%, similar to a control with kaolin (10–20%), but lower than bentonite or celite (30–45%). The observations reported here, although inconclusive as to the function of geophagy in this species, indicate that it is not a mineral supplement during times of scarcity or high consumption of leaves, as has been reported for other primates, nor that it is related to fruit consumption (redressing possible mineral imbalance), as has been suggested for some other frugivorous mammals. Our results do not rule out tannin adsorptive hypothesis for the ingestion of clays, but, being an irregular habit, we argue that it is most likely related to rare and occasional dietary components.