Because of their strong economic impact, complexity and maintainability are among the most widely used terms in software engineering. But, they are also among the most weakly understood. A multitude of software metrics attempts to analyze complexity and a proliferation of different definitions of maintainability can be found in text books and corporate quality guide lines. The trouble is that none of these approaches provides a reliable basis for objectively assessing the ability of a software system to absorb future changes. In contrast to this, relational database theory has successfully solved very similar difficulties through normal forms. In this paper, we transfer the idea of normal forms to code. The approach taken is to introduce semantic dependencies as a foundation for the definition of code normal form criteria