The development of tools like calcium imaging, electrophysiology, and optogenetics now allows the recording and manipulation of neural activity in behaving animals 1, 2, 3, 4. Yet, there are no comparable tools for studying behavior, and progress in systems and behavioral neuroscience has lagged behind progress in cellular and molecular neuroscience. Explanations of this state of affairs include limitations of current experimental designs [5] and the lack of statistical rigor [6], but a critical reason has been overlooked — namely the inadequacy of behavioral measures commonly used in neuroscience experiments. With the recent development of methods for quantifying continuous behavior with high temporal and spatial resolution 7, 8, 9, 10, the technical limitations are now being overcome, yet conceptual confusions still remain. There is no consensus on basic questions like the function of neural signals and the nature of neural coding 11, 12, 13, 14.

To understand the limitations of conventional approaches in the study of behavior, it is necessary to examine their implicit assumptions. Here, I first examine three assumptions that have misled researchers who try to understand the relationship between brain and behavior: 1) behavior is output, 2) behavior is a discrete event, and 3) behavior is the effect of some cause. I then review some recent results that support the feasibility of an alternative approach and discuss their implications for studying the relationship between brain and behavior.