It is thought that the first continental crust formed by melting of either eclogite or amphibolite, either at subduction zones or on the underside of thick oceanic crust. However, the observed compositions of early crustal rocks and experimental studies have been unable to distinguish between these possibilities. Here we show a clear contrast in trace-element ratios of melts derived from amphibolites and those from eclogites. Partial melting of low-magnesium amphibolite can explain the low niobium/tantalum and high zirconium/samarium ratios in melts, as required for the early continental crust, whereas the melting of eclogite cannot. This indicates that the earliest continental crust formed by melting of amphibolites in subduction-zone environments and not by the melting of eclogite or magnesium-rich amphibolites in the lower part of thick oceanic crust. Moreover, the low niobium/tantalum ratio seen in subduction-zone igneous rocks of all ages is evidence that the melting of rutile-eclogite has never been a volumetrically important process.