Numbers are highly relevant in our everyday lives. Besides intentionally processing number magnitude when necessary, we often automatically process it even when not required. The SNARC (Spatial-Numerical Association of Response Codes; Dehaene et al., 1993) effect, describing faster left-/right-sided responses to smaller/larger numbers, respectively, provides evidence for this automaticity. It arises in semantic number-processing tasks both when number magnitude is task-relevant (e.g., magnitude classification) and task-irrelevant (e.g., parity judgment). However, findings on the SNARC effect in tasks requiring the processing of non-semantic number features are mixed: while it has been observed in orientation judgment tasks, it was mostly absent in color judgment tasks. Importantly, previous studies were underpowered or did not control for confounding variables. In two highly powered online experiments, we found a small but significant SNARC effect in both nominal color judgment (cyan vs. yellow; slope = 1.71 ms) and color intensity judgment (light cyan vs. dark cyan; slope = 1.13 ms) of Arabic digits from 1 to 9 excluding 5, which did not significantly differ in size. Further, we found little evidence for the MARC (Linguistic Markedness of Response Codes, i.e., faster left-/right-sided responses to odd/even numbers, respectively; Nuerk et al., 2004) effect. Moreover, the odd effect (i.e., faster responses to even than to odd numbers; Hines, 1990) was detected. Taken together, both magnitude and parity are processed automatically even if participants respond to physical non-semantic and non-spatial number features, but the spatial mapping seems more automatic for magnitude than for parity.