Many current efforts to develop large-scale science assessments involve hands-on tasks because of their presumed power to elicit and measure scientific reasoning skills. An analysis of the processes in which students engage while responding to such assessment is needed in order to discover the specific forms of reasoning that tasks elicit. This paper describes a framework that organizes the cognitive demands that assessment tasks place on students. The framework is applied to a set of science tasks completed by 20 sixth-grade students using think-aloud protocols, observations, and interviews. This procedure revealed several ways in which tasks required skills not anticipated by the test developers and provided a richer understanding of what successful performance entailed. Three tables and one figure present study findings. (Contains 32 references.) (Author/SLD)