The stream power law, expressed as E = KAmSn — where E is erosion rate [LT − 1], K is an erodibility coefficient [T − 1L (1 − 2m)], A is drainage area [L 2], S is channel gradient [L/L], and m and n are constants — is the most widely used model for bedrock channel incision. Despite its simplicity and limitations, the model has proved useful for topographic evolution, knickpoint migration, palaeotopography reconstruction, and the determination of rock uplift patterns and rates. However, the unknown parameters K, m, and n are often fixed arbitrarily or are based on assumptions about the physics of the erosion processes that are not always valid, which considerably limits the use and interpretation of the model. In this study, we compile a unique global data set of published basin-averaged erosion rates that use detrital cosmogenic 10Be. These data (N = 1457) enable values for fundamental river properties to be empirically constrained, often for the first time, such as the concavity of the river profile (m/n ratio or concavity index), the link between channel slope and erosion rate (slope exponent n), and substrate erodibility (K). These three parameters are calculated for 59 geographic areas using the integral method of channel profile analysis and allow for a global scale analysis in terms of climatic, tectonic, and environmental settings. In order to compare multiple sites, we also normalise n and K using a reference concavity index m/n = 0.5. A multiple regression analysis demonstrates that intuitive or previously demonstrated local-scale trends, such as the correlation between K and precipitation rates, do not appear at a global scale. Our results suggest that the slope exponent is generally > 1, meaning that the relationship between erosion rate and the channel gradient is nonlinear and thus support the hypothesis that incision is a threshold controlled process. This result questions the validity of many regional interpretations of climate and/or tectonics where the unity of n is routinely assumed.