To effectively reduce illegal timber trade, law enforcers need forensic methods to independently verify claims of wood origin. Multi-element analysis of traded plant material has the potential to be used to trace the origin of commodities, but for timber it has not been tested at relevant large scales. Here we put this method to the test, by evaluating its tracing accuracy for three economically important tropical timbers: Azobé and Tali in Central Africa (22 sites) and Red Meranti on Borneo (9 sites). Wood samples from 991 trees were measured using Inductively Coupled Plasma Mass Spectrometry and element concentrations were analysed to chemically group similar sites (clustering) and assess accuracy of tracing samples to their origin (Random Forest models). For all three timbers, we found distinct spatial differences in chemical composition. In Central Africa, tracing accuracy was 86%-98% for regional clusters of chemically similar sites, with accuracy depending on the tracing question. These clusters were 50-800 km apart and tracing accuracy was highest when combining the two timbers. Tracing accuracy of Red Meranti on Borneo was 88% at the site level. This high accuracy at a small scale may be related to the short distances at which differences in soil type occur on Borneo. A blind sample analysis of 46 African timber samples correctly identified the origin of 70%-72% of the samples, but failed to exclude 70% of the samples obtained from different species or outside the study area. Overall, these results illustrate a high potential for multi-element analysis to be developed into a timber tracing tool which can identify origin for multiple species and can do so at a within-country scale. To reach this potential, reference databases need to cover wider geographic areas and represent more timbers.