Selective attention has been studied extensively using the negative priming (NP) paradigm. An important issue regards the representational level at which NP occurs. We investigated this issue by using numbers as stimuli. Because numbers have a well-defined semantic organization, which can be clearly measured by means of the distance effect, they are very suitable for testing the assumption that NP is situated at a central semantic level. Four experiments are presented in which the numerical distance between prime distractor and probe target was manipulated. The task was magnitude comparison. Target and distractor were defined on the basis of colour. In Experiment 1, all numbers were presented in Arabic format; NP was observed only with identical prime distractor and probe target, and no distance-related NP was observed. This could not be explained by a decay of inhibition since in Experiment 2 similar results were obtained with a shortened response-to-stimulus interval. Experiment 3 showed that these observations also hold for numbers presented verbally. Nevertheless, a cross-notational experiment with Arabic prime and verbal probe (Experiment 4) revealed no NP whatsoever and excluded the possibility that the absence of distance-related negative priming was the result of a fine-tuned inhibitory mechanism operating at the semantic level. The results are considered in the light of current theories of negative priming.