A matrix is presented of the errors of perception made by 135 men and women listening to three male and three female speakers reading aloud different randomized lists constructed from the letters of the alphabet and the digits 1—9, heard in white noise. Data from a short-term memory (STM) experiment, using simultaneous visual presentation and immediate ordered recall of two selected vocabularies of nine letters and the digits 1-9, are cited as evidence of phonemic confusion between letters and digits in STM. Conrad (1964) established a high correlation between the systematic errors made by Hsteners identifying letters of the alphabet spoken one at a time in white noise (auditory confusions), and those made by subjects when visually presented with strings of alphabetic material for immediate ordered recall. Briefly, items which sounded similar were more probably confused not only in listening, but also in short-term memory (STM). Once this relationship between auditory and STM errors had been demonstrated, it was possible, using Clarke's constant ratio rule (Clarke, 1957), to select from the listening matrix subsets of letters of diifering probabilities of auditory confusion. Conrad {\&} Hull (1964) have shown that such subsets selected for minimal auditory confusability, whether of three or nine items in the vocabulary, are significantly better recalled, after visual sequential presentation, than subsets of letters of high auditory confusability. In STM tasks using both letters and digits, Wickelgren (1965) with auditory pre-sentation, and Hintzman (1965) using visual stimuli, have shown that inter-class errors are also related to phonemic similarity. Definition of the extent and relative confusability of letters with digits has been hampered by the lack of a complete matrix of auditory confusions amongst these stimuli. The provision of such a matrix was the purpose of this study. METHOD AND PEOCEDTTKB