It is thought that continental crust existed as early as 150 million years after planetary accretion1, but assessing the rates and processes of subsequent crustal growth requires linking the apparently contradictory information from the igneous and sedimentary rock records. For example, the striking global peaks in juvenile igneous activity 2.7, 1.9 and 1.2 Gyr ago imply rapid crustal generation in response to the emplacement of mantle 'super-plumes', rather than by the continuous process of subduction 2-4. Yet uncertainties persist over whether these age peaks are artefacts of selective preservation5, and over how to reconcile episodic crust formation with the smooth crustal evolution curves inferred from neodymium isotope variations of sedimentary rocks6,7. Detrital zircons encapsulate a more representative record of igneous events than the exposed geology1,8,9 and their hafnium isotope ratios reflect the time since the source of the parental magmas separated from the mantle. These 'model' ages are only meaningful if the host magma lacked a mixed or sedimentary source component10, but the latter can be diagnosed by oxygen isotopes, which are strongly fractionated by rock-hydrosphere interactions. Here we report the first study that integrates hafnium and oxygen isotopes, all measured in situ on the same, precisely dated detrital zircon grains. The data reveal that crust generation in part of Gondwana was limited to major pulses at 1.9 and 3.3 Gyr ago, and that the zircons crystallized during repeated reworking of crust formed at these times. The implication is that the mechanisms of crust formation differed from those of crustal differentiation in ancient erogenic belts.