The present thesis models young children’s vocabulary acquisition using different semantic network structures in hopes of better understanding the underlying cognitive representations children have about the words they know. Throughout each chapter, we predicted the specific word-by-word learning of toddlers using words’ centrality in different network representations. First, we found that gathering semantic similarity using distributional methods from a child-directed language corpus better predicts specific learning than using an adult language corpus. We then investigated three different semantic representations, the same distributional method, word associations, and feature norms. Each semantic type had both a child-based and adult-based version. We found modest confirmation that child-based sources of semantic similarity are indeed better than adult-based sources, and further found that association- based semantics best capture the knowledge young children have and use to learn new vocabulary. The current work lays a foundation for understanding the underlying structure of young children’s vocabulary knowledge.