It has long been understood that associations can form between items that are paired (Ebbinghaus, 1885), but it is commonly assumed that previously retrieved items are not used when remembering items in serial order. We present a series of experiments that test this assumption, using a serial learning procedure inspired by Ebenholtz (1963). In this procedure, participants practiced recalling ordered lists of letters, and the order of the letters was manipulated. Half of the lists were scrambled such that the serial positions and relative positions of the letters were inconsistent over practice. The other half of the lists were spun (e.g., ABCDEF [right arrow] FABCDE), making the serial positions inconsistent but preserving the relative positions of the letters over practice. In Experiment 1, participants learned to recall spun lists more accurately than the scrambled lists with practice. In Experiments 2 and 3, participants recalled new lists more accurately when they shared the relative order of previously learned spun lists. In Experiments 4 and 5, the influences of motor and perceptual representations were removed and shown to have little impact on the advantage for spun lists. Experiment 6 extended our findings to the more traditional Hebb (1961) learning procedure. The results of our experiments indicate that the commonly held assumption is incorrect--previously retrieved items can contribute to memory for serial order. Previously retrieved items best serve serial memory when there is ample opportunity to strengthen item-to-item associations.