Test-Driven Development (TDD) is often touted as a best practice for software development. However studies of the impact of TDD in traditional software development have so far shown mixed results in practice. This paper is one of the few to examine experience with TDD specifically in the scientific computing area. Based on a community survey, it analyzes the benefits reported by users of TDD, the problems reported in adopting TDD, and challenges and solutions of refactoring (an important part of the TDD process). Although the authors' assessment is that, once again, the impacts of TDD are mixed, the more detailed discussion of the findings should make this paper a useful reading for anyone considering implementing TDD in their development processes. Not only might it help decide whether or not to proceed with TDD, but it might also help those who proceed adapt the process to obtain better results.