Having students identify for themselves where they think they need to improve and then comparing this to their performance on a task will give an accurate insight into which parts of the criteria they do not understand how to satisfy. The teacher can then point this out and explain how it can be achieved. A good method to conduct this is by using 'predictive grids':
Too often, students see the grade that they receive and do not take enough notice into the written feedback that is given during marking. Dylan Wiliam's suggestion of not giving the grade to the students works but I've been wanting the students to engage with the success criteria when it comes to their exams and they really are incredibly motivated by seeing the grades, or even a number.
Something suggested by a colleague, which I've started to implement with exam classes, is 'predictive grids'. These are success criteria grids (usually using the exam board language as much as possible) which the students highlight according to what they think they achieved in the piece of work. This can either be done directly after the piece of work is finished or the lesson after which is what I usually prefer to do. When the work is marked, the teacher can then see where specifically the students are not achieving, but also see if the students know if they are not achieving in that particular area.