The effects of item position (the serial position effect: SPE) on the recognition of a list of arms presented in an 8-arm radial maze have not been shown to be robust across studies which differ in numbers of subjects, amount of training and task difficulty. The present study examined whether more robust SPEs could be obtained in rats with a 12-arm radial maze using a matching-to-sample serial probe recognition (SPR) procedure. In Part 1, 23 rats received extensive training on recognition of list of arms from a 5-arm list and, in Part 2, 20 rats received training with a 7-arm list. Both Parts 1 and 2 showed that a reliable and persistent SPE emerged due to superior recognition of items at the first and last serial positions compared with the middle positions. However, the SPE was more pronounced with the 7-arm than 5-arm list. It is argued that adequate task difficulty along with sufficient subject and trial numbers is necessary to produce a clear SPE. In Part 3, an alternative means of assessing accuracy derived from signal-detection theory is examined, and the procedural requirements for its use are identified. Although this measure did not alter the conclusions reached using percent correct, it is proposed that the bias-free measure, log d, is superior to traditional indices of memory performance which may obscure the presence of, or changes in, the SPE. The procedures used here provide a valuable means to produce clear, reliable and persistent SPEs. Such procedures are essential if researchers are to be confident about the effects of lesions or drugs upon list memory in their attempts to explore the neurological bases of memory and model human neurological disorders.