Modelling reading aloud is a special case in the growing ﬁeld of computational cognitive modelling. The SM model is indeed one of the most successful computational models of cognitive functions: this is testiﬁed by the huge number of citations, and by the major debate in the reading literature generated by the model. While in the past years the success of a computational model could be judged by simple “proof of simulation”, this has now quite changed. A good model should account for data from experimental studies on normal subjects (in terms of both reaction times and error rates), from studies on patients with neuropsychological disorders, and ultimately from developmental studies. However, the strengths of a model must be judged not only in terms of number of facts that it can account for, but also in terms of economy and relative “transparency” of the model: if the model’s internal operations are diﬃcult to interpret, little insight can be gained from it (for discussion, see McCloskey, 1991). A further important point regards the novel or diﬀerential predictions that models may provide.