Introduction​: Semantic long-term memory (LTM) representations can be distinguished in two main classes: thematic (i.e., concrete, context-dependent) and taxonomic (i.e., more abstract, space-time independent; see Blaye et al. 2001; Mandler et al. 1987). In typically developing children, taxonomies are usually acquired after thematic representations; an understudied topic, we addressed here, was to investigate how LTM semantic representations may modulate WM recall in atypically developing children, in particular children with developmental dyslexia. Methods​: A sample of 66 dyslexic children (mean age 10.69 years; 45 males) balanced with a control group for age, gender, schooling and IQ, was administered a semantic WM task (see Belacchi et al. 2017) in order to collect recall accuracy and possible intrusions. Here, children, had to listen to groups of lists composed of words semantically associated (thematic, e.g. light-heat-fire, or taxonomic, e.g., shop-drugstore-cof ee) or arbitrarily associated, and afterwards to recall the last words among each group. Results​: Both taxonomic and thematic associations supported recall (compared to arbitrary associations) in the two groups of children. More specifically, data showed that in typically developing children the taxonomic association boosted WM recall (vs. the thematic one). On the contrary, in dyslexic children the taxonomic association did not prevail over the thematic one. Discussion​:Dyslexic children performed poorly (compared to control) on a task requiring recall of semantically associated words. Whereas in typical development taxonomies favour LTM organization and WM performance, in developmental dyslexia taxonomies do not. Results were interpreted in the light of a possible lack of integration between intelligence and language in developmental dyslexia. References​: Belacchi, C et al. Ps Cl Svil (2017). 1:159-172. Blaye, A et al. Brit J Dev Psych (2001). 19:395-412. Mandler JM et al. Cognit Dev (1987). 2:339-354.