A 'relational agent' is a computer agent designed to build and maintain a long-term social-emotional relationship with a user. In this talk I will describe my work in developing and evaluating such agents, including communication strategies these agents can use to establish trust, liking and working alliance with users, and applications in which this is particularly useful for improving task outcomes. Since people build relationships primarily in the context of face-to-face conversation, and studies have demonstrated that nonverbal aspects of human interaction are most crucial during social (as opposed to purely task-oriented) interactions, I have been investigating the use of embodied conversational agents (ECAs) as the medium for doing this work. ECAs are animated computer characters that use many of the nonverbal channels of communication found in human face-to-face interaction-including hand gesture, gaze cues, facial display, head nods, and body posture, in order to provide more effective and natural interfaces. Another dimension of my research is the development of technologies to support long-term, repeated interaction with agents, including the use of persistent memory and strategies for referring back to previous interactions and maintaining engagement with users over time. This work has primarily been applied in the area of health behavior change, and I will present results from two randomized controlled intervention studies that I have conducted in the area of physical activity promotion. I'll also spend some time talking about ongoing projects and plans for the future.