The modular framework of number processing (e.g., S. Dehaene & R. Akhavein, 1995) was applied to study sequential trial-to-trial effects in a number comparison task. In Experiment 1, numbers were always presented as digits. Responses were faster when the same number was repeated, but this effect was additive with the numerical distance effect. In Experiment 2, numbers were presented either as digits or as words. The authors found significant effects of repeating (a) the same physical stimulus, (b) the same number but in a different notation, and (c) the same notation but a different number. Again, all 3 effects were additive with the numerical distance effect. The authors' results provide strong evidence against accounts according to which, on stimulus repetition trials, the comparison stage is bypassed (as proposed by S. Dehaene, 1996), and the results clearly favor an early, precomparison locus of repetition effects.