Studies of the hydroclimate at regional scales rely on spatial rainfall data products, derived from remotely-sensed (RS) and in-situ (IS, rain gauge) observations. Because regional rainfall cannot be directly measured, spatial data products are biased. These biases pose a source of uncertainty in environmental analyses, attributable to the choices made by data-users in selecting a representation of rainfall. We use the rainforest-savanna transition region in Brazil to show differences in the statistics describing rainfall across nine RS and interpolated-IS daily rainfall datasets covering the period of 1998-2013. These differences propagate into estimates of temporal trends in monthly rainfall and descriptive hydroclimate indices. Rainfall trends from different datasets are inconsistent at river basin scales, and the magnitude of index differences is comparable to the estimated bias in global climate model projections. To address this uncertainty, we evaluate the correspondence of different rainfall datasets with streamflow from 89 river basins. We demonstrate that direct empirical comparisons between rainfall and streamflow provide a method for evaluating rainfall dataset performance across multiple areal (basin) units. These results highlight the need for users of rainfall datasets to quantify this "data selection uncertainty" problem, and either justify data use choices, or report the uncertainty in derived results.