When interacting autonomous agents operate in open domains that contain novel objects, the agents may be unable to refer to the objects by using preassigned names. As an alternative, the agents might use descriptions. When agents refer to domain objects by using descriptions, referent misidentifications (RMIs) will occur. This paper formally models recovery from RMIs among such agents. We show how the RMI recovery problem can be cast in a belief revision framework and how recovery can be effected with relatively simple mechanisms. For instance, neither agent explicitly reasons about the beliefs of others.