We critically evaluate the potential and limitations of an alternative way to calculate erosion rates based on petrographic and mineralogical fingerprints of fluvial sediments coupled with gauged sediment fluxes. Our approach allows us to apportion sediment loads to different lithological units, and consequently to discriminate erosion rates in different tectonic domains within each catchment. Our provenance data on modern Taiwanese sands indicate focused erosion in the Backbone Range and Tananao Complex of the retrowedge. Lower rates are inferred for the northern part of the island characterized by tectonic extension and for the western foothills in the prowedge. The principal factor of uncertainty affecting our estimates is the inevitably inaccurate evaluation of total sediment load, because only the suspended flux was measured. Another is the assumption that suspended load and bed load are derived from the same sources in fixed proportions. Additional errors are caused by the insufficiently precise definition of lithologically similar compositional end-members and by the temporal variability of sediment composition at the outlet of each catchment related to the spatial variability of erosional processes and triggering agents such as earthquakes, typhoons, and landslides. To evaluate the robustness of our findings, we applied a morphometric technique based on the stream-power model. The results obtained are broadly consistent, with local discrepancies ascribed to poorly constrained assumptions and choices of scaling parameters. Our local erosion estimates are consistent with GPS uplift rates measured on a decadal timescale and generally higher than basin-wide results inferred from cosmogenic-nuclide and thermochronology data.