When subjects study a list of B items paired with A items and are test
ed for recognition of the B items alone, and then for recall of the B
items given the A items as cues, B items that can be recalled frequent
ly go unrecognised. The extent of this recognition failure is predicta
ble from a function discovered by Tulving and Wiseman (1975), which re
lates the probability of recognising the recallable items to the proba
bility of recognising all items. Two kinds of exceptions to this funct
ion have been discovered: encoding exceptions and retrieval exceptions
. Very few observations of retrieval exceptions exist. Four experiment
s described in this article provide further evidence that such excepti
ons occur when the B items are categorised and the A items are the nam
es of the categories. According to a contextual account of the functio
n and exceptions to it, these retrieval exceptions occur because the i
nformation provided by the A-item cues can be largely retrieved from t
he B items in the recognition test; hence, as cues, the A items do not
provide much different contextual information to that provided by the
B items. By this account, the function is an empirical law; exception
s fall outside the range of this law and define its boundary condition
s.