I argue that an adequate understanding of the practice of constructing models in physics requires a distinction between two strategies that are commonly both labeled ‘idealization’. The formal characteristic of both methods is to let a parameter in the equations for a target system go to zero. But the discussion of examples from various applications of perturbation theory shows that there is in general a difference with respect to the aims such limiting procedures are supposed to serve; and with different aims comes the need to characterize the means (the interpretation of the limits) differently. I therefore suggest that we distinguish ‘idealizations’ from ‘perspectives’ or perspectival representations.