This paper reports on a study where the incidental memory of 18 children with ADHD and 18 typically developing peers was assessed by means of a conventional two-phase recognition memory test. In the study phase participants were required to categorize as a living or non-living a set of 64 stimuli from 8 semantic categories. In the test phase, they were required to recognize "target" (i.e., stimuli from the first set) from "non-target" stimuli. Children with ADHD were overall less accurate and much slower than TD controls in identifying both living and non-living items. Moreover, while most of TD participants made very few, if any, errors, only 7 out of 18 participants with ADHD scored near ceiling, and 2 of them scored below chance level. Following the Signal Detection Theory approach, the participants' performance on the test phase was scored in terms of d prime (d') values. Children with ADHD had lower d' indexes compared to controls both for living and non-living stimuli, although this difference did not reach statistical significance. More interestingly, the variability of the d' values was higher in the ADHD compared to Controls. Taken together, findings from this study indicate that at least some of the children with ADHD have a genuine impairment in processing visual stimuli. More generally, these results cast doubts on the idea that ADHD represents a stable nosographic entity.