In many parts of Africa, soil erosion is an important problem, which is evident from high sediment yields in tropical montane streams. Previous studies in Kenya pointed to a large contribution from catchments cultivated by smallholder farmers. This led to the hypothesis that unpaved tracks and gullies are the main sediment sources in smallholder agriculture catchments of the highlands of Kenya. The aim of this study was to investigate the sediment sources with sediment fingerprinting to generate the knowledge base to improve land management and to reduce sediment yields. Four main sediment sources (agricultural land, unpaved tracks, gullies and channel banks) and suspended sediments were analysed for biogeochemical elements as potential tracers. To apportion the catchments target sediment to different sources, we applied the MixSIAR un-mixing modelling under a Bayesian framework. Surprisingly, the fingerprinting analysis showed that agricultural land accounted for 75% (95% confidence interval 63–86%) of the total sediment. Channel banks contributed 21% (8–32%), while the smallest contributions to sediment were generated by the unpaved tracks and gullies with 3% (0–12%) and 1% (0–4%), respectively. Erosion management strategies should target agricultural lands with an emphasis on disconnecting unpaved tracks form hillslope source areas to reduce sediment yields to Lake Victoria.