Test-Driven Development is a software development technique based on quick cycles that switch between writing tests and implementing a solution that assures that tests do pass. Test-Driven Development has produced excellent results in various aspects of building new software systems. Increased maintainability, improved design, reduced defect density, better documentation and increased code test coverage are reported as advantages that contribute to reducing the cost of development and, consequently, to increasing return on investment. All these benefits have contributed for Test-Driven Development to become an increasingly relevant practice while developing software. When evaluating test driven development from the perspective of maintaining legacy systems, we face a clear mismatch when trying to adopt this technique. Test-Driven Development is based on the premise that tests should be written before coding, but when working with legacy code we already have thousands of lines written and running. Considering this context, we discuss in this dissertation a technique, which we call Test-Driven Maintenance, that is a result of adapting Test-Driven Development to the needs of maintaining legacy systems. We describe how we have performed the adaptation that lead us to this new technique. Afterwards, we evaluate the efficacy of the technique applying it to a realistic project. To obtain realistic evaluation results, we have performed an empirical study while introducing the technique in a maintenance team working on a legacy system that is in constant evolution and use by an enterprise in Rio de Janeiro.