In this paper we propose a paradigm called the Interactive Visual Dialog (IVD) as a means of facilitating a system’s ability to recognize objects presented to it by a human. The presentation centers around a supermarket checkout scenario in which an operator presents an item to be tallied to a stationary television camera. An active vision approach is used to provide feedback to the operator in the form of an image (or images) depicting what the system thinks the operator is most likely holding, shown in a viewpoint that suggests how the object should next be presented to improve the certainty of interpretation. Interaction proceeds iteratively until the system converges on the correct interpretation. We show how the IVD can be implemented using an entropy-based gaze planning strategy and a sequential Bayes recognition system using optical flow as input. Experimental results show that the system does, in practice, improve recognition accuracy, leading to convergence to a correct solution in a minimal number of iterations. In this paper, we explore how a visual dialog between a person and a machine can be used to facilitate interpretive tasks such as recognizing objects. The presentation will center around a supermarket checkout scenario in which an operator sweeps an object to be identified in front of a stationary television camera. Instead of a bar code, the system must recognize the object from the sequence of images generated as a result of its motion in front of the camera. This task is difficult because imaging conditions cannot be precisely controlled (e.g. object pose, distance to camera, illumination, etc.) so that it is likely that the system will fail to correctly identify the object in a significant number of instances. In previous work, [1] we showed how an active vision approach could be used to solve a similar recognition problem in the context of a mobile robot in a stationary environment. There, ambiguity of recognition in the form of entropy measures was used to calculate gaze trajectories that minimized the uncertainty of interpretation. The present problem is more difficult for two reasons i) motion is induced by a human instead of a robot and ii) the requested motions must somehow be communicated to the human. Since the variability of human motions cannot be controlled, they must be treated as noise and accounted for by the recognition process. Rather than attempting to base recognition on a single measurement, a hypothesis filtering strategy is applied to the entire sequence so that evidence for different object hypotheses can be accumulated over time