A strong magnetic circular dichroism has been observed in core-level photoemission in a geometry where it is considered to be forbidden-that is when the X-ray wave vector and the sample magnetization vector are perpendicular. This observation is in sharp contradiction to existing models of circular magnetic dichroism. Recent experiments studying the 2p levels of iron indicate that a consistent interpretation of all the results is obtained when the final-state-selection effects of angle-resolved photoemission are taken into account.