49 páginas, 13 figurasOver the last century extraordinary efforts have been devoted to determining soil erosion rates (in units of mass per area and time) under a large range of climatic conditions and land uses, and involving various measurement methods. We undertook a meta-analysis of published data from more than 4000 sites worldwide. The results show that there is extraordinarily high variability in erosion rates, with almost any rate apparently possible irrespective of slope, climate, scale, land use/land cover and other environmental characteristics. However, detailed analysis revealed a number of general features including positive relationships of erosion rate with slope and annual precipitation, and a significant effect of land use, with agricultural lands yielding the highest erosion rates, and forest and shrublands yielding the lowest. Despite these general trends, there is much variability that is not explained by this combination of factors, but is related, at least partially, to the experimental conditions. Our analysis revealed a negative relationship between erosion rate and the size of the study area involved; significant differences associated with differing measurement methods, with direct sediment measurement yielding the lowest erosion rates, and bathymetric, radioisotope and modeling methods yielding the highest rates; and a very important effect of the duration of the experiment. Our results highlight that, when interpreting erosion rates, the experimental conditions involved must be taken into account. Even so, the data suggest that only order of magnitude approximations of erosion rates are possible, and these retain a very large degree of uncertainty. Consequently, for practical purposes such as calculation of global sediment budgets, empirical equations are not a substitute for direct measurements. Our results also show that a large proportion of the experiments have been short-term (less than 3. years), which reduces dramatically the reliability of the estimated erosion rates, given the highly non-normal behavior of soil erosion (time-dependency). Despite the efforts already made, more long-term measurement experiments need to be performed, especially in regions of the world that are under-represented in global datasets. In addition, protocols need to be established for standardizing the measurement methods and reporting the results, to enable data to be compared among diverse sites.Support for this research was provided by the projects INDICA (CGL2011-27753-C02-01 and -02) and HIDROCAES (CGL2011-27574-C02-C01), funded by the Spanish Ministry of Economy and Competitiveness, and an agreement between the CSIC and the Spanish Ministry of Environment (RESEL). The Geomorphology and Global Change research group was financed by the Aragón Government and the European Social Fund (ESF-FSE) (E68). Estela Nadal-Romero and Yasmina Sanjuán were the recipients of a “Marie Curie-IEF” postdoctoral contract (Project 624974) and an FPI pre-doctoral contract from the European Commission and the Spanish Ministry of Economy and Competitiveness, respectively.Peer reviewe