Abstract: With the development of social media, influencers and virtual influencers have become one of the most important channels for customers to learn about brands and products, and have a profound impact on consumers' brand attitudes, purchase intentions, etc. However, how to choose the right influencers, especially influencers with suitable facial features, has become the primary challenge for marketing practitioners. However, we have limited understanding of the impact of influencers' faces, especially virtual influencers' faces on customers' decisions to follow influencers. Therefore, we applied the hierarchical linear multinomial logit model to quantitatively analyze the relationship between influencers' facial features and customers' decisions to follow influencers, and study whether the effects of facial features of real and virtual influenceis are different. The results show that ： (1) models based on facial features can accurately explain and predict consumers' preferences for influencers' faces; (2) when faces of virtual influencers are indistinguishable from faces of real ones, there aren't any systematic differences between the effects of facial features on the consumers' preferences. Theoretically, the paper fills the research gap on the relationship between facial features and consumers' following behaviors in influencer marketing, especially in virtual influencer marketing ； practically, our findings can help guide influencer marketing practices, especially the selection and design of faces of influencers.

Key words: influencer marketing , virtual influencers , face image analytics , hierarchical linear multinomial logit model