A new model of recognition memory is reported in its simplest form. The model is applied both to basic findings and some additional phenomena that pose problems for current models of recognition memory. These problematic results include the list-strength effect (e.g.Ratcliff, Clark & Shiffrin, 1990). Simulations demonstrate that the model is capable of predicting these phenomena in a correct qualitative manner. The model assumes item and context to consist of a vector of feature values, and assumes storage of separate episodic images, each an incomplete and error prone copy of the studied vector. The probability that a test item is old' is calculated, and a default old' response given if this probability is greater than .5.