Barberousse, Franceschelli and Imbert claim that ''simulations can be used as experiments because they represent phenomena'', their arguments being based on a detailed analysis of the semantic levels of computer simulations. We give a computer scientist point of view of this analysis, making it even more detailed and relate it to work on formal semantics in computer science. We argue that the use of formal methods applied to computer simulations make possible the precise assessment of the accuracy of simulations, making scientists more confident on their outcomes.