The phonological similarity effect--poor retention of order for lists of similar-sounding items--is a benchmark finding in the short-term memory literature. In our first two experiments, we show that the effect actually reverses following relatively brief periods of distraction, yielding better order retention for similar than for dissimilar lists, provided that different items are used on every trial. In Experiment 3, the same items were used on every trial and similar lists produced poorer performance across all three retention intervals. The results are interpreted from a general discrimination framework: Items are viewed as occupying positions in a multidimensional space defined by list and within-list dimensions.