Some simulationists have argued that the information obtained during the perceptual process of facial expression (the geometric features) is sufficient for recognition of the emotion intended by that expression. Drawing on evidence from cross-cultural studies, with particular attention to conceptual act theories, I show that both emotion expression and recognition are top-down modulated by expressivity norms, observer-specific internal representations, and expectations. I thus conclude that direct simulation, or a purely bottom-up approach, is not sufficient for emotion recognition. Next, I will (...) consider the generate and test strategy as an alternative to direct simulation. This is a highly promising approach in that top-down and bottom-up processes work in parallel, according to the presence, or not, of additional information. For this reason, it avoids the difficulties of direct simulation. I will argue, however, that it is based on a misconception of emotions, specifically on how emotions are represented in the brain. To improve the model, I distinguish between two different categorical approaches to emotions, then argue that only one of which is viable in a simulational account of emotion recognition. (shrink)