COMETS are extremely fragile objects. Observations of outbursts and splitting have inspired suggestions that their nuclei are ’rubble piles’, consisting of components that are weakly bonded, perhaps held together by mutual gravity(1,2). This structure was confirmed in spectacular fashion by comet Shoemaker-Levy 9, which disintegrated into about 20 fragments after passing near Jupiter; the tidal stresses induced by the planet were only of the order of 10(-4) bar (ref. 3). Attempts to explain the formation of comets in the outer Solar System have emphasized either collisional coagulation(1) or gravitational collapse of a layer of dust particles(4,5). Here I argue that the observed sized and structure of comet nuclei are better explained by a two-stage process, involving elements of both models-collisional coagulation in the solar nebula, followed by gravitational instability of a layer of macroscopic bodies.