Emotional facial expressions are immediate indicators of the affective dispositions of others. Recently it has been shown that early stages of social perception can already be influenced by (implicit) attributions made by the observer about the agent's mental state and intentions. In the current study possible mechanisms underpinning distortions in the perception of dynamic, ecologically-valid, facial expressions were explored. In four experiments we examined to what extent basic perceptual processes such as contrast/context effects, adaptation and representational momentum underpinned the perceptual distortions, and to what extent 'emotional anticipation', i.e. the involuntary anticipation of the other's emotional state of mind on the basis of the immediate perceptual history, might have played a role. Neutral facial expressions displayed at the end of short video-clips, in which an initial facial expression of joy or anger gradually morphed into a neutral expression, were misjudged as being slightly angry or slightly happy, respectively (Experiment 1). This response bias disappeared when the actor's identity changed in the final neutral expression (Experiment 2). Videos depicting neutral-to-joy-to-neutral and neutral-to-anger-to-neutral sequences again produced biases but in opposite direction (Experiment 3). The bias survived insertion of a 400 ms blank (Experiment 4). These results suggested that the perceptual distortions were not caused by any of the low-level perceptual mechanisms (adaptation, representational momentum and contrast effects). We speculate that especially when presented with dynamic, facial expressions, perceptual distortions occur that reflect 'emotional anticipation' (a low-level mindreading mechanism), which overrules low-level visual mechanisms. Underpinning neural mechanisms are discussed in relation to the current debate on action and emotion understanding.