Data from two studies were used to estimate the reliability of facial EMG when used to index facial mimicry (Study 1) or affective reactions to pictorial stimuli (Study 2). Results for individual muscle sites varied between muscles and depending on data treatment. For difference scores, acceptable internal consistencies were found only for corrugator supercilii, and test-retest reliabilities were low. For contrast measures describing patterns of reactions to stimuli, such as high zygomaticus major combined with low corrugator supercilii, acceptable internal consistencies were found for facial reactions to smiling faces and positive affective reactions to affiliative images (Study 2). Facial reactions to negative emotions (Study 1) and facial reactions to power and somewhat less to achievement imagery (Study 2) showed unsatisfactory internal consistencies. For contrast measures, good temporal stability over 24 months (Study 1) and 15 months (Study 2), respectively, was obtained. In Study 1, the effect of method factors such as mode of presentation was more reliable than the emotion effect. Overall, people's facial reactions to affective stimuli seem to be influenced by a variety of factors other than the emotion-eliciting element per se, which resulted in biased internal consistency estimates. However, the influence of these factors in turn seemed to be stable over time.