Spatial–numerical associations are observed when participants perform number categorization tasks. One such observation is the spatial numerical associations of response codes (SNARC) effect, showing an association between small numbers and the left-hand side and between large numbers and the right-hand side. It has long been argued that this spatial association is automatically activated by the long-term representation underlying numbers processing. Instead, van Dijck and Fias (2011) argued that this association is a short-term representation that is constructed during task execution. This argument was based on the observation of an association between the ordinal position of an item in working memory and response side (e.g., the ordinal position effect). Four different experiments were set up to systematically investigate this assumption. Our results indicate that the activation of the canonical order of numbers in working memory (e.g., 1, 2, 3, etc.) is indeed necessary to observe the SNARC effect. The activation of the standard sequence of numbers (e.g., from 1 to 9) can be overruled when a new random sequence is memorized. However, this is only observed when retrieval of the memorized sequence is required during the numbers classification task.