In Phase 1, a total of 313 hypermedia features were articulated, based on the four exercises students engaged in while working with the IBM Ultimedia programs Columbus and Illuminated Books and Manuscripts. Nine domains emerged from the 313 responses ; these domains were further divided into 29 subdomains. The dominant domain was some aspect of Learner Control which was divided into three primary subdomains : Pace, Navigation, and Choice. Whereas Pace can be attributed to any computer-based program (which places hypermedia programs within the more encompassing term of computer-based learning), Navigation and especially Choice are clearly features of hypermedia environments. Although there were some general Navigation features given, the majority of them dealt with Nonlinear Navigation, a professed benefit of hypermedia. The three main types of Choice were (a) information based on a concept or a theme, (b) information that enhanced the context of a concept or theme that was already selected, and (c) information (concept-based or context-enhancing) that could come in various media forms. Phase 2 detailed the evolution of the 313 hypermedia features. An important finding in this phase was that general hypermedia features tended to occur early in the students' use of the Ultimedia programs : features such as Pace, Navigation, and Choice. However, later in their experience with the Ultimedia programs, the features students articulated tended to be more detailed and specific (type of Choices and how the information should be stored, or programmed, in the environment). The students' experience with the hypermedia environment could be viewed as top-down-or general to specific-which reflects a sound approach to learning. Finally in Phase 3, the features stated by three students with various magnitudes of attitude changes were compared. The person with the least hypermedia experience had the greatest change in attitude toward hypermedia. Despite her greater attitudinal change, when compared with the person who had the least attitudinal change, the features she provided were, almost without exception, based on some form of Learner Control. Such a finding suggests that this participant was never able to view the hypermedia environment except as a user of it. The person who changed the least attitudinally, but also had more experience with hypermedia programs, stated features that were predominantly not from a Learner Control perspective. She was able to view the hypermedia environment from various viewpoints and angles. Collectively, these three phases have some implications for instruction about, or in, hypermedia environments. Students are conscious of the numerous, varied features of hypermedia environments, with a clear predominance of Learner-Control features. The students' awareness of such features increases in complexity as they experience the environment more. Another point is that students with differing backgrounds in hypermedia and with varying attitudes toward the environments tend to view the features very differently. Instructional implications can be summed up by three points. First, expecting students to be more critical of hypermedia environments may require systematic instructional exercises so that students do not view these environments from only a Learner-Control perspective. Second, moving students from articulating simple to complex features takes time.