Transpiration is a major component of evapotranspiration (ET), critically needed in modeling ecosystem functions and dynamics at large scales.  While catchment water balance (i.e., precipitation minus runoff) can provide coarse-scale information about this process, tree-based estimates of catchment transpiration based on sap flux from heat dissipation probes may provide an important alternative.  We performed uncertainty analysis of Monte Carlo simulation to examine error propagation associated with this bottom-up approach to estimating catchment transpiration from probed sapwood to tree to watershed, using data from a homogeneous watershed in Coweeta Basin.  We analyzed six uncertainty sources: (1) measurement error of probe data; (2) variability in tree variables (e.g., diameter, sapwood thickness, crown radius); (3) original calibration error from Granier’s empirical relationship for sap flux; (4) radial profile integration error from estimating sap flow in the unprobed sapwood area; (5) time integration error from summing 15-minute sap flow values to annual estimates; and (6) conversion error from sap flow to canopy transpiration.  Our objectives were to quantify uncertainty in catchment transpiration estimates, determine relative contributions of error sources, and explore ways of reducing uncertainty in the scaling process.  Results showed that the maximum uncertainty in transpiration estimates was high with CV of 78%, that variability in tree variables was the most critical factor, accounting for 48% of the estimated variance of transpiration, and that CV of transpiration estimates may be reduced by 50% via averaging of multiple trees within plots.  With information obtained here, better scaling strategies can be developed for heterogeneous landscapes.