Ecohydrological processes in tropical rainforests are insufficiently understood, and existing studies yield contradictory results. We investigated relative contributions of different soil depths to tree water uptake of 83 trees and possible species‐specific differences in a 50 × 50 m forest plot at four dates in a tropical montane forest in Kenya using stable water isotopes and the Bayesian mixing model framework MixSIAR. We found distinct individual tree differences (e.g. Drypetes gerrardii taking 75% of its water from <0.5 m, or a rather large shift in uptake patterns based on the climatic conditions, that is the fourth sampling date), but no consistent species‐specific or small‐scale spatiotemporal patterns in water uptake and depth contributions. Soil water δ18O showed a lateral variation of up to 6‰, which was accounted for by a spatial interpolation of soil water isotopes and enabled us to improve allocations of water uptake sources to individual trees. Our results show that ignoring the lateral variability of water isotope signatures in soils complicates the applicability of a mixing model in this context and might be a widespread constraint reducing the validity and comparability of mixing model results. Further research on underlying processes of water fluxes in forest ecosystems is urgently needed and we point out the need for considering large individual differences in water uptake patterns and small‐scale variability of soil water isotopic composition despite homogeneous soil characteristics.