This paper studies the relationship between egocentric collaboration networks and knowledge creation at the individual level. For egocentric networks we focus on the characteristics of tie strength and tie configuration, and knowledge creation is assessed by the number of citations. Using a panel of 1042 American scientists in five disciplines and fixed effects models, we found an inverted U-shaped relationship between network average tie strength and citation impact, because an increase in tie strength on the one hand facilitates the collaborative knowledge creation process and on the other hand decreases cognitive diversity. In addition, when the network average tie strength is high, a more skewed network performs better because it still has a "healthy" mixture of weak and strong ties and a balance between exploration and exploitation. Furthermore, the tie strength skewness moderates the effect of network average tie strength: both the initial positive effect and the later negative effect of an increase in tie strength are smaller in a more skewed network than in a less skewed one.