This work analyzes the behavior and effectiveness of a science team using the Marsokhod mobile robot to explore the Silver Lake region in the Mojave Desert near Baker, California. The work addresses the manner in which the geologists organized themselves, how they allocated their time in different activities, how they formed and communicated scientific hypotheses, and the frequency with which they requested different types of data from the mission archive during the first 3 days of the mission. Eleven scientists from the NASA Ames Research Center and three of the five scientists who participated from their home institutions were videotaped as they worked throughout the 3-day experiment. The videotape record indicates that 46% of available person-hours were consumed in semistructured or formal meetings and that only 1% of their time was spent studying immersive, three-dimensional virtual reality models of the robot's surroundings. The remainder of their time was spent in unstructured work sessions in groups of two or three. Hypothesis formation and evolution patterns show a meager flow of information from the distributed science team to the on-site team and a bias against reporting speculative hypotheses. Analysis of the visual imagery received from the robot indicates that acquisition of the large panoramic information leads to high levels of redundancy in the data acquired. The scientists' archive requests indicate that small, specifically requested image targets were the most frequently accessed information. The work suggests alternative organizational structures that would expedite the flow of information within the geologic team. It also advocates emphasizing specific science targets over high-resolution, stereoscopic, panoramic imaging when programming a mobile robot's onboard cameras. ¿ 2001 American Geophysical Union