Teaching and learning about abstraction

It is a commonplace in computer science education that abstraction is a key concept. In model-driven development in particular, we all understand that the power of models comes from the fact that a model is an abstract representation of something important about a system. As a community, we are sometimes conflicted about whether the ability to think abstractly - to generate "good" abstractions, and/or to use abstractions effectively - is innate, or can be taught. To whatever extent it can be taught, how is this best done? I will review the state of the art concerning these matters, relating it to my experience in teaching modelling to university students, and to my brief trials of teaching it to ten-year-old school children.