This paper presents formative Web testing and evaluation methods created at the Exploratorium, ranging from floor testing with museum visitors using low-cost ëguerrillaí methods to structured evaluations that engage museum visitors and on-line remote audiences in the design process. This paper illustrates these methods using examples from two U.S. National Science Foundation-funded informal science education projects: "Evidence: How Do We Know What We Know?" addresses how to improve the publicís understanding of a key element of the scientific process: how scientists attempt to construct a functional understanding of the world by gathering evidence, and how they make discoveries based on evidence. The second project, and "Mind: Attention, Emotion, and Judgment -- How do minds figure out what to do?" addresses developing a companion Web site to a new exhibition and educational programs that support museum visitors' exploration and experimentation with their own minds. In both of these projects, close collaboration between Web developers and evaluators enabled the collection of useful feedback. Different configurations of interactive media, the Internet, the museum floor, and visitors were used in concert. to disentangle open-design questions and to generate constructive feedback and Web design revisions. This partnership approach and these evaluation methods offer developers new alternatives to standard usability testing approaches for improving Web site designs in the early development stages of complex projects.