How do Java methods grow?

Overly long methods hamper the maintainability of software - they are hard to understand and to change, but also difficult to test, reuse, and profile. While technically there are many opportunities to refactor long methods, little is known about their origin and their evolution. It is unclear how much effort should be spent to refactor them and when this effort is spent best. To obtain a maintenance strategy, we need a better understanding of how software systems and their methods evolve. This paper presents an empirical case study on method growth in Java with nine open source and one industry system. We show that most methods do not increase their length significantly; in fact, about half of them remain unchanged after the initial commit. Instead, software systems grow by adding new methods rather than by modifying existing methods.