Experiments are described in which tin was diffused into GaAs in a closed ampoule. When the ambient arsenic vapour pressure was low the results were not reproducible, and severe deterioration of the surface was observed. Evidence is produced which suggests that both evaporation and condensation occurred during these diffusions. The problem was completely solved by employing a high-vapour pressure of arsenic. It is suggested that a gallium-rich surface is less stable than one which is arsenic-rich.