The tidy classification system that divided γ-ray bursts (GRBs) into long-duration busts (lasting more than two seconds) and short may have had its day. The final nail in its coffin may be GRB 060614. Discovered on 14 June 2006 by the Burst Alert Telescope on-board the Swift satellite, this burst was long, at 102 seconds, but as reported in a clutch of papers in this issue, it has a number of properties, including the absence of an accompanying supernova, that were previously considered diagnostic of a 'short' GRB. The hunt is now on for a classification system to take account of the diversity now apparent in GRBs. In the accompanying News & Views, Bing Zhang suggests that the answer may be to adopt a Type I/Type II classification similar to that used for supernovae. Optical observations of γ-ray burst (GRB) 060614 (duration ∼100s) rule out the presence of an associated supernova. This would seem to require a new explosive process: either a massive 'collapsar' that powers a GRB without any associated supernova, or a new type of engine, as long-lived as the collapsar but without a massive star. Over the past decade, our physical understanding of γ-ray bursts (GRBs) has progressed rapidly, thanks to the discovery and observation of their long-lived afterglow emission. Long-duration (≳2 s) GRBs are associated with the explosive deaths of massive stars (‘collapsars’, ref. 1), which produce accompanying supernovae2,3,4,5; the short-duration (≲2 s) GRBs have a different origin, which has been argued to be the merger of two compact objects6,7,8,9. Here we report optical observations of GRB 060614 (duration ∼100 s, ref. 10) that rule out the presence of an associated supernova. This would seem to require a new explosive process: either a massive collapsar that powers a GRB without any associated supernova, or a new type of ‘engine’, as long-lived as the collapsar but without a massive star. We also show that the properties of the host galaxy (redshift z = 0.125) distinguish it from other long-duration GRB hosts and suggest that an entirely new type of GRB progenitor may be required.