Ensuring usability has become a key goal of interactive system development, as developers have begun to realize that it matters little how effectively an interactive system can compute, if human users cannot communicate effectively with the system. In this paper we discuss what we have found to be two main types of formative user interface evaluation: analytic and empirical. Both these types occur as part of the development process. We do not attempt to survey all approaches to either of these types of formative evaluation, but rather to offer a sampling of some approaches that have been found (by us and by others) to be useful in ensuring usability. We give only an overview of analytic methods, and then focus on empirical methods. We conclude with some observations on future trends in user interface evaluation.