Abstract Uranium-series isotopes can be used to determine constraints on the timescale of slab dehydration and melt production at subduction zones. However, interpretations of U–Th–Ra data suggest very different timescales of slab dehydration. Here, we present new U–Th–Ra data from Kamchatka along with a number of alternative models for production of radioactive disequilibrium. Variations in ( 226 Ra/ 230 Th) and ( 231 Pa/ 235 U) activity ratios are best explained by crystal fractionation with host rock assimilation for a duration of less than c. 6000 years. The association of the largest 226 Ra excesses with high Sr/Th in the most primitive lavas suggests that Ra–Th fractionation is controlled by slab dehydration less than 10 ka ago. We show that U–Th data can be explained by dynamic melting of a recently (<10 ka) metasomatized mantle wedge. Dynamic melting of an oxidized source metasomatized several hundreds of thousands of years ago cannot produce significant 231 Pa excess. Because 238 U– 230 Th disequilibrium is inferred to be controlled by partial melting, there is no requirement for multi-stage slab dehydration commencing ∼150 ka. We suggest that Ra–Th disequilibria constrain the timing of slab dehydration, whilst U–Th fractionation is dominated by partial melting, at least at the Kamchatka arc.