In cognitive neuroscience, functional magnetic resonance imaging (fMRI) is used to produce images of brain functions. These images play a central role in the practice of neuroscience. In this paper we are interested in how these brain images become understandable and meaningful for scientists. In order to explore this problem we observe how scientists use such semiotic resources as gesture, language, and material structure present in the socially and culturally constituted environment. A micro-analysis of video records of scientists interacting with each other and with fMRI images reveals action as cognition, that is, actions that constitute thinking for the scientists.