We show in this paper that the analysis of diffusion-induced instability in spatially extended models can be performed by separating local dynamics from diffusion. This is possible not only in the case studied by Turing, namely models with two interacting variables, but also in the general case of three or more variables. The advantage of this decomposition, based on the notion of potential Turing instability, is illustrated through the analysis of two spatially extended plant-insect models.