Evidence is summarized that suggests that when a protogalaxy collapses, a fraction $f$ of its gas fails to heat to the virial temperature, where $f$ is large for haloes less massive than the value $M^*$ associated with $L^*$ galaxies. Stars and galaxies form only from the cool gas fraction. Hot gas is ejected from low-mass systems as in conventional semi-analytic models of galaxy formation. In high-mass systems it is retained but does not cool and form stars. Instead it builds up as a largely inert atmosphere, in which cooling is inhibited by an episodically active galactic nucleus. Cold gas frequently falls into galactic haloes. In the absence of a dense atmosphere of virial-temperature gas it builds up on nearly circular orbits and forms stars. When there is a sufficiently dense hot atmosphere, cold infalling gas tends to be ablated and absorbed by the hot atmosphere before it can form stars. The picture nicely explains away the surfeit of high-luminosity galaxies that has recently plagued semi-analytic models of galaxy formation, replacing them by systems of moderate luminosity from old stars and large X-ray luminosities from hot gas.

5 pages. Version to appear in MNRAS Minor changes & corrected bibliography since first version