Generally, people respond faster to small numbers with left-sided responses and large numbers with right-sided responses, a pattern known as the SNARC (spatial numerical association of response codes) effect. The SNARC effect is interpreted as evidence for amodal automatic access of magnitude and its spatial associations, because it occurs in settings where number is task-irrelevant and for different number formats. We report five studies designed to establish the degree to which activation of magnitude and its spatial associations is truly automatic and amodal. Based on the notion of autonomous automaticity, we hypothesized that the mere presence of a number form (to which participants made a color decision) would be sufficient to elicit the SNARC effect. However, we found no evidence of a SNARC effect for simple color decisions to Arabic digits (Experiment 1). There was a SNARC effect for color decision to digits when participants recognized the stimulus as a digit before responding (Experiment 2), participants viewed the digit for sufficient time before color onset (Experiments 3 and 5), or there was temporal uncertainty regarding color onset (Experiment 3). There was no SNARC effect for color decision to arrays of circles (Experiment 4), regardless of viewing time or temporal uncertainty. Overall, our results suggest that, while access to magnitude and its spatial associations is not automatic in an "all-or-none" sense, it is certainly at the strong end of automaticity, and that this automatic activation is modality dependent. Our findings are most supportive of conceptual coding accounts of the SNARC effect.