This study investigates whether mimicry of facial emotions is a stable response or can instead be modulated and influenced by memory of the context in which the emotion was initially observed, and therefore the meaning of the expression. The study manipulated emotion consistency implicitly, where a face expressing smiles or frowns was irrelevant and to be ignored while participants categorised target scenes. Some face identities always expressed emotions consistent with the scene (e.g., smiling with a positive scene), whilst others were always inconsistent (e.g., frowning with a positive scene). During this implicit learning of face identity and emotion consistency there was evidence for encoding of face-scene emotion consistency, with slower RTs, a reduction in trust, and inhibited facial EMG for faces expressing incompatible emotions. However, in a later task where the faces were subsequently viewed expressing emotions with no additional context, there was no evidence for retrieval of prior emotion consistency, as mimicry of emotion was similar for consistent and inconsistent individuals. We conclude that facial mimicry can be influenced by current emotion context, but there is little evidence of learning, as subsequent mimicry of emotionally consistent and inconsistent faces is similar.