Over the last 5 years we have developed, produced, tested, and evaluated an authoring software package to produce web-based, interactive, audio-enhanced language-learning material. That authoring package has been used to produce language-learning material in French, Spanish, German, Arabic, and Tamil. We are currently working on increasing accessibility by pooling digital resources and a new project to develop a server-based online assessment package. Having run this software development process through five versions of the authoring package and updated the language material three times, we feel in a position to comment of the application of software lifecycle theory to the specific needs of CALL research and development. Starting from a brief outline of classical software development lifecycle theory, the paper will go through the practical experience gained by our own five iterations, using examples of our work to suggest how the standard theoretical framework can be adapted to the specific needs of CALL. In particular, the fact that there are distinct sets of users of CALL software--teachers and learners--affects the approach. Different sets of requirements have to be balanced and, indeed, even within a group there are differing requirements as teachers can be authors of new material, users of the end product, or both. Ours is a small team that has been involved in all stages of the software development process as well as subsequently using the resulting package to author our own material, using it in the teaching process, and learning new languages ourselves using material produced by our system. This has given valuable and unusual insight into the whole process. On occasions this formal approach has been neglected and the work has sometimes suffered as a result. It will be argued that the computer industry standard approach to modelling, the Unified Modelling Language (UML), is particularly applicable to CALL development and should be more widely used. (Contains 3 figures.)