The development of high level programming knowledge is not well known and there are no direct techniques to evaluate and measure it. Measurement of programming skills is mostly based on the evaluation of programs written by students but the final programs do not reveal the correctness of students' mental models of programming concepts. This paper presents a new analysis model that can be used to evaluate students' mental models based on their program writing protocols with three different measures. The analysis model is based on the concept of roles of variables and extends the methodology of Rist's theory of schema expansion in programming but has a better coverage and is more finely grained than Rist's model. This paper also reports a study in which the applicability and the general validity of the model was studied by following the emergence of novice students' programming knowledge during their first programming course. The results suggest that the model is generally applicable for analyzing programming knowledge in novice level and that the three measures can be applied when focusing on different aspects of this knowledge.