Attention deficit hyperactivity disorder (ADHD) is characterized by deficient self-regulation, poor attentional control, and poor response inhibition. To date, however, the extent to which these deficits affect basic interference control remains a matter of controversy. Secondly, ADHD has been reported to be associated with arithmetic deficits. It remains unclear whether such deficits are a secondary consequence of the above-mentioned characteristics of ADHD or whether basic numerical magnitude representations are also affected. In the present study we attempted to investigate these issues using a basic numerical interference paradigm. Nine- to twelve-year-old children with ADHD-C (attention-deficit hyperactivity-disorder combined type) and control children without ADHD (each n = 16) were presented with two digits of possibly different physical sizes (e.g., 3 7). This numerical Stroop task requires subjects to make a magnitude classification concerning either the physical or the numerical stimulus dimension. The irrelevant dimension can be congruent (same response), incongruent (different response), or neutral (no response association). Children with ADHD-C performed worse than control children in most analyses. The most important finding was a significant interaction of congruity effects with group in the numerical comparison task. Children with ADHD-C tended to show larger congruity and interference effects than controls, and these were not attributable to a speed-accuracy trade-off. The results might reflect differential processing speeds, or a different degree of automatic activation of physical and numerical magnitudes in children with and without ADHD-C. Alternative explanations, such as insufficient inhibition of selective (domain-specific) attention are also discussed. The order of authorship is alphabetical, as the authors contributed equally to the study. We wish to thank the participating children, their parents, and their schools for contributing to this project. We are also grateful to Susanna Bloder and Maria Hoellwarth for data collection, and Elise Klein and Katharina Dressel for checking the references. In particular, we wish to thank Stuart Fellows for checking English style and grammar. This research was supported by funding of the European Community (Neuromath: HPRN-CT-2000-00076 and RTN NUMBRA proposal 504927), by grants of the RWTH Aachen (IZKF “BioMAT.”; VV N50, 51, 69c; START AZ 160/05) and the German Research Society (DFG: KFO110/TP2) supporting the research of Hans-Christoph Nuerk.