<p>In elementary school classrooms, Computational Thinking (CT) teaching often is embedded in problem- or project-based pedagogies and involves programming using Educational Robotics. Little is known, however, about how children solve problems in these contexts, and how task designs influence which CT and problem-solving processes they apply. In this study, we examined children’s problem-solving and enactment of CT processes in two different task contexts. We analyzed talk and actions of 16 pairs of 8- and 9-year-old children who programmed robots in distinct task contexts: a more structured context (a simulation task) vs. a more open-ended context (a path-finding task). Results showed that in the path-finding task children used increasingly lower-level problem-solving steps and more frequently relied on trial-and-error approaches. A contrasting profile was found in the simulation task. We link the contrasting profiles to task designs and contextual factors. We call for empirical studies of classroom learning activities that enable children to create models of CT teaching, so as to support instructional designers and teachers in the development of task designs aligned with instructional goals and children’s problem-solving and CT skills.</p>