In an increasingly dynamic business environment, innovation output occupies a central position among all organizational outputs, not only because it is a primary way in which firms compete and grow, but also because it profoundly influences social and economic evolution (Eisenhardt & Tabrizi, 1995; Sorensen & Stuart, 2000). Understanding the factors that determine an organization’s ability to produce new ideas and continually innovate thus is a fundamental issue in strategic management and marketing fields. Among all sort of determinants, collaboration with buyers to create value through innovation has attracted particular attention from scholars (von Hippel, 1988; Thomke & von Hippel, 2002; Sawhney, Verona, & Prandelli, 2005). This study examined the role of Internet-based collaboration in the buyer-supplier relationship in promoting product innovation of supplying firms. Drawing on the juxtaposition of the governance literature and social exchange theory, we proposed that Internet-based collaboration positively affects product innovation performance of supplying firms, but too much dependence on it impedes product innovation. That is, Internet-based collaboration has an inverted U-shaped relationship with product innovation performance of supplying firms. It further posited that in-person interaction between buying and supplying firms strengthens the positive effect of Internet-based collaboration on product innovation, such that when the degree of in-person interaction is high, Internet-based collaboration is associated with better innovation performance. These propositions were tested using data from a large survey conducted by a survey data on buyer-supplier relationships in China. The results show that Internet-based collaboration has a stronger positive relationship with innovation performance and this positive effect declines after the degree of Internet-based collaboration goes beyond a threshold. Moreover, the first-order effect of Internet-based collaboration is stronger when the degree of in-person interaction is high than when it is low. The optimal level of Internet-based collaboration in low degree of in-person interaction is moderate, whereas when in-person interaction is more frequently used in connecting with buying firms the optimal level rises. These results provide strong supports for the predictions of hypotheses.