From the earliest ages at which infants search for hidden objects, they make the A (B) over bar error, searching perseveratively at previous rather than current hiding locations (Piaget, 1954). This paper presents a parallel distributed processing (PDP) model that instantiates an explicit set of processing mechanisms to account for a large and diverse set of data on infants' A (B) over bar errors. The model demonstrates how basic processes U the formation of latent memory traces and their interaction with developing active memory traces U can provide a unifying framework for understanding why and when infants perseverate. Novel predictions from the model are discussed, together with its challenges for theories that posit a concept of object permanence in the first year of life.