Abstract A model of dike emplacement at divergent plate boundaries is developed which predicts that below the uppermost 1–3 km of the oceanic crust, the number and lengths of dikes in any particular swarm should increase, but that the thickness should decrease, with depth in the crust. These predictions are supported by field observations. It is proposed that rapid injections of dikes at divergent plate boundaries may temporarily alter the normal stress field in such a way that, at a certain level in the crust, subsequent dikes change into sills. Such sills may develop into crustal magma chambers, either when many nearby sills combine into a large one, or when an initial sill absorbs the magma of many dikes that enter it in a rapid succession while the sill is liquid. For the fastest spreading ridges the initial thickness of a sill that is to develop into a crustal chamber need only attain 20 m, whereas at the slowest spreading ridges the initial thickness must exceed 170 m. Because thin sills are much more likely to form than thick sills, the probability of sills developing into crustal chambers increases with dike intrusion frequency, hence with spreading rate. Furthermore, the model predicts, first, that with increasing spreading rate the length of the chamber along the ridge axis should increase, and, second, that the most likely site for magma chambers is at depths of 2–4 km. Both these predictions are supported by available data.