The question is approached through three examples. Ross's model, although very simple and formulated a priori, yielded important epidemiplogical insights: the existence of a threshold contact rate (vectorial capacity); the decreasing sensitivity of the endemic level to changes in the contact rate, as the latter gets larger; the return to the same equilibrium endemic level, as long as the contact rate remains the same; the progressively decreasing impact of a given reduction in the contact rate until a new equilibrium is reached. The second example is Macdonald's model, in particular his sensitivity analysis; two constraints are pointed out: the weakest point, on which to concentrate control efforts, cannot be identified automatically by the sensitivity analysis; the calculation of the expected impact of an intervention commonly assumes too much uniformity (e.g. of human of vector behaviour) and this commonly leads to exaggerated expectations. The third example is the Garki model, briefly considered in terms of its assumptions, of its behaviour, of its actual utilization (only for teaching, so far), and of the cost of its development. Looking forward, three uses of models are discussed. It is suggested that, critically used, they have a place in training, in planning control and in research. With respect to their application to planning, it is argued that the need for new data is not necessarily great, also that some rather difficult direct measurements might be substituted by indirect measurements, a point illustrated by the expected relationships, following the Garki model, between different dimensions of “intensity” of malaria. With respect to research, it is suggested that simulations may assist in clarifying discussions around the expected impact of malaria vaccines, hence guide their field testing.