A thorough understanding of the changes in rainfall extremes in the past is needed for making reliable projections in the future. These past extremes can be investigated using gridded observations of daily rainfall. Despite the availability of several observationally constrained datasets of daily precipitation based on rain gauge measurements, remote sensing and/or reanalyses, we demonstrate a large disparity in the global land mean of daily precipitation intensity. Surprisingly, the magnitude of this spread is similar to that found in CMIP5 models. Even when comparing interpolated data based on the same network of in situ stations, there exists uncertainty due to varying interpolation methods. For example in Australia, our results point to distinct structural uncertainties between products derived solely from in situ observations (interpolated datasets) and two products that combine both remote sensed data and in situ observations. We also show a large spread in the upper quantiles between the various datasets compared, indicating that substantial uncertainty exists in extremes from various gridded datasets. These large uncertainties within observations critically underpin our understanding of rainfall extremes in the past and in turn projections into the future.