It is argued that current models of memory do not adequately account for the confabulations that are found in the recall of certain neurological patients. A model of the relation between control processes and memory involved in recalling autobiographical episodes is put forward. It is based on detailed analysis of the protocols of healthy volunteers' autobiographical recollections of recent everyday events. It is held that damage to different components of the model fits with the different patterns of performance found in confabulators, and examples of the errors that confabulators make are discussed in terms of those made by normal subjects.