The fate of subducted sediment and the extent to which it is dehydrated and/or melted before incorporation into arc lavas has profound implications for the thermo-mechanical nature of the mantle wedge and models for crustal evolution. In order to address these issues, we have undertaken the first measurements of 10Be and light elements in lavas from the Tonga–Kermadec arc and the sediment profile at DSDP site 204 outboard of the trench. The 10Be/9Be ratios in the Tonga lavas are lower than predicted from flux models but can be explained if (a) previously estimated sediment contributions are too high by a factor of 2–10, (b) the top 1–22 m of the incoming sediment is accreted, (c) large amounts of sediment erosion are proposed, or (d) the sediment component takes several Myr longer than the subducting plate to reach the magma source region beneath Tonga. The lavas form negative Th/Be–Li/Be arrays that extend from a depleted mantle source composition to lower Th/Be and Li/Be ratios than that of the bulk sediment. Thus, these arrays are not easily explained by bulk sediment addition and, using partition coefficients derived from experiments on the in-coming sediment, we show that they are also unlikely to result from fluid released during dehydration of the sediment (or altered oceanic crust). However, partial melts of the dehydrated sediment residue formed at ∼800 °C during the breakdown of amphibole±plagioclase and in the absence of cordierite have significantly lowered Th/Be ratios. The lava arrays can be successfully modelled as 10–15% partial melts of depleted mantle after it has been enriched by the addition of 0.2–2% of these partial melts. Phase relations suggest that this requires that the top of the subducting crust reaches temperatures of ∼800 °C by the time it attains ∼80 km depth which is in excellent agreement with the results of recent numerical models incorporating a temperature-dependent mantle viscosity. Under these conditions the wet basalt solidus is also crossed yet there is no recognisable eclogitic signal in the lavas suggesting that on-going dehydration or strong thermal gradients in the upper part of the subducting plate inhibit partial melting of the altered oceanic crust.