Here we examine the question of the chemical models widely used to describe dense solutions, particularly ionic solutions. First, a simple macroscopic analysis shows that, in the case of weak interactions, taking into account aggregated species amounts to modelling an effective attraction between solutes, although the stoichiometry used does not necessarily correspond to atomic reality. We then use a rigorous microscopic analysis to explain how, in the very general case, chemical models can be obtained from an atomic physical description. We show that there are no good or bad chemical models as long as we consider exact calculations. To obtain the simplest possible description, it is nevertheless advisable to take the speciation criterion that minimises the excess terms. Molecular simulations show that, very often, species can be defined simply by grouping ions which are in direct contact. In some cases, the appearance of macroscale clusters can be predicted.