Deuterium-labeled water was used to study the effect of the Tapajós Throughfall Exclusion Experiment (TTEE) on soil moisture movement and on depth of water uptake by trees of Coussarea racemosa, Sclerolobium chrysophyllum, and Eschweilera pedicellata. The TTEE simulates an extended dry season in an eastern Amazonian rainforest, a plausible scenario if the El Niño phenomenon changes with climate change. The TTEE excludes 60% of the wet season throughfall from a 1-ha plot (treatment), while the control 1-ha plot receives precipitation year-round. Mean percolation rate of the label peak in the control plot was greater than in the treatment plot during the wet season (0.75 vs. 0.07 m/mo). The rate was similar for both plots during the dry season (ca. 0.15 m/mo), indicative that both plots have similar topsoil structure. Interestingly, the label peak in the control plot during the dry season migrated upward an average distance of 64 cm. We show that water probably moved upward through soil pores-i.e., it did not involve roots (hydraulic lift)-most likely because of a favorable gradient of total (matric + gravitational) potential coupled with sufficient unsaturated hydraulic conductivity. Water probably also moved upward in the treatment plot, but was not detectable; the label in this plot did not percolate below 1 m or beyond the depth of plant water uptake. During the dry season, trees in the rainfall exclusion plot, regardless of species, consistently absorbed water significantly deeper, but never below 1.5-2 m, than trees in the control plot, and therefore may represent expected root function of this understory/subcanopy tree community during extended dry periods.