A substantial increase or decrease, relative to a control condition, in rate of acquisition of a paired-associate list was demonstrated when formal similarity between stimulus and response items was located within and between pairs, respectively. In addition, the results show that the effect of formal intrapair similarity can be modified by the predictability of the repeated letter. The location of the within-pair repeated letter does not affect the total number of errors but it does affect the distribution of the types of errors made by S.