It is important to have an accurate student model when attempting to give remedial teaching with an Intelligent Tutoring Systems (ITS). The Leeds Modeling System (LMS) uses a predefined set of rules and mal-rules (incorrect variants observed in pupils’ protocols) to define a space of possible student models. If the predefined set is not sufficient, the generating of further mal-rules is necessary. The paper describes a schema to generate mal-rules. Some heuristics are used in this procedure. There is also a certain relationship with formal languages, grammars, and parsing.