Interactive robots are used increasingly not only in entertainment and service robotics, but also in rehabilitation, therapy and education. The work presented in this paper is part of the Aurora project, rooted in assistive technology and robot-human interaction research. Our primary aim is to study if robots can potentially be used as therapeutically or educationally useful ‘toys’. In this paper we outline the aims of the project that this study belongs to, as well as the specific qualitative contextual perspective that is being used. We then provide an in-depth evaluation, in part using Conversation Analysis (CA), of segments of trials where three children with autism interacted with a robot as well as an adult. We focus our analysis primarily on joint attention which plays a fundamental role in human development and social understanding. Joint attention skills of children with autism have been studied extensively in autism research and therefore this behaviour provides a relevant focus for our study. In the setting used, joint attention emerges from natural and spontaneous interactions between a child and an adult. We present the data in the form of transcripts and photo stills. The examples were selected from extensive video footage for illustrative purposes, i.e. demonstrating how children with autism can respond to the changing behaviour of their co-participant, i.e. the experimenter. Furthermore, our data shows that the robot provides a salient object, or mediator for joint attention. The paper concludes with a discussion of implications of this work in the context of further studies with robots and children with autism within the Aurora project, as well as the potential contribution of robots to research into the nature of autism.