Traditional accounts of verbal short-term memory explain differences in performance for different types
of verbal material by reference to inherent characteristics of the verbal items making up memory
sequences. The role of previous experience with sequences of different types is ostensibly controlled
for either by deliberate exclusion or by presenting multiple trials constructed from different random permutations.
We cast doubt on this general approach in a detailed analysis of the basis for the robust finding
that short-term memory for digit sequences is superior to that for other sequences of verbal material.
Specifically, we show across four experiments that this advantage is not due to inherent characteristics of
digits as verbal items, nor are individual digits within sequences better remembered than other types of
individual verbal items. Rather, the advantage for digit sequences stems from the increased frequency,
compared to other verbal material, with which digits appear in random sequences in natural language,
and furthermore, relatively frequent digit sequences support better short-term serial recall than less frequent
ones. We also provide corpus-based computational support for the argument that performance in a
short-term memory setting is a function of basic associative learning processes operating on the linguistic
experience of the rememberer. The experimental and computational results raise questions not only
about the role played by measurement of digit span in cognition generally, but also about the way in
which long-term memory processes impact on short-term memory functioning.