Clonal plantations of Eucalyptus have been introduced since 1978 on savanna soils of the coastal plains of Congo. Atmospheric deposition, canopy exchange and transfer through the soil were estimated on the whole rooting depth (6 m) over 3 years, in an experimental design installed in a native savanna and an adjacent 6-year-old Eucalyptus plantation. Complementary measurements after planting the experimental savanna made it possible to establish input-output budgets of nutrients for the whole Eucalyptus rotation and to compare them with the native savanna ecosystem. In this highly-weathered soil, atmospheric deposits and symbiotic N fixation by a legume species balanced the nutrient budgets in savanna, despite large losses during annual burnings. After afforestation, weeding in the Eucalyptus stands eliminated the leguminous species responsible for a N input by symbiotic fixation of about 20 kg ha-1 year-1. Whereas the budgets of P, K, Ca and Mg were roughly balanced, the current silviculture led to a deficit of about 140 kg N ha-1 in the soil, throughout a 7-year rotation. This deficit was large relative to the pool of total N in the upper soil layer (0-50 cm), which was about 2 t ha-1. Therefore, the sustainability of Congolese plantations will require an increase in N fertilizer inputs over successive rotations to balance the N budget. These results were consistent with field trials of fertilization. Practical consequences of these budgets were identified, in order to: (i) direct field trials of fertilization, (ii) select appropriate methods of soil preparation, weed control and harvest, (iii) highlight the importance of fire prevention in this area, and (iv) support the implementation of field trials aiming at introducing a biological nitrogen fixing understorey in Eucalyptus stands.