When people solve or construct mathematical word problems, they tend to align semantic relations between object sets with analogous mathematical relations (Bassok, Chase, & Martin, 1998). In particular, people tend to add taxonomically related objects (e.g., tulips and daffodils) but refrain from adding thematically related objects (e.g., tulips and vases). To examine whether such "semantic alignments" are automatic, we adapted Lefevre and Kulak's (1994) number-matching task in which subjects were briefly exposed to a digit pair (e.g., 3+5) followed by a single digit probe. In rejecting probes that did not match either number in the digit pair, subjects took longer to reject a sum probe (e.g., 8) than a neutral probe (e.g., 9). We combined digit pairs with words denoting object sets to examine whether automatic activation of addition facts (e.g., 3+5=8) is stronger when the addends are aligned with taxonomically related objects (e.g., 3 tulips + 5 daffodils) than with unrelated objects (e.g., 3 tulips + 5 cars). We found some evidence for co-activation of semantic and arithmetic relations and also uncovered methodological issues to be explored in future studies.