A growing body of scholarly work indicates that student performance on physics problems stems from many factors, including relevant conceptual understanding. However, in contexts in which significant conceptual difficulties have been documented via research, it can be difficult to pinpoint and isolate such factors because students' written and interview responses rarely reveal the full richness of their conscious and, perhaps more importantly, subconscious reasoning paths. In this investigation, informed by dualprocess theories of reasoning and decision making as well as the theoretical construct of accessibility, we conducted a series of experiments in order to gain greater insight into the factors impacting student performance on the "five-block problem," which has been used in the literature to probe student thinking about buoyancy. In particular, we examined both the impact of problem design (including salient features and cueing) and the impact of targeted instruction focused on density-based arguments for sinking and floating and on neutral buoyancy. The investigation found that instructional modifications designed to remove the strong intuitive appeal of the first-available response led to significantly improved performance, without improving student conceptual understanding of the requisite buoyancy concepts. As such, our findings represent an important first step in identifying systematic strategies for using theories from cognitive science to guide the development and refinement of research-based instructional materials.