As premodern texts become increasingly accessible in digital form, a major obstacle to their computational analysis is the lack of structured resources for identifying and normalizing historical entities. This challenge is especially evident in Classical Chinese, where standardized benchmarks for end-to-end entity linking remain limited. To address this need, we present ACEL-Eval, a benchmark dataset for evaluating entity linking in Classical Chinese. ACEL-Eval is derived from the Shiji and covers six entity categories that frequently occur in classical historical writing: person , location , state , official title , book , and time . The dataset supports end-to-end entity linking by combining named entity recognition over six entity categories with entity disambiguation for person mentions. To facilitate person referent normalization, we provide a person-oriented knowledge base containing core attributes such as personal names, aliases, and state affiliation. We conduct baseline experiments on the two target tasks to validate dataset usability and to establish reproducible reference results for future comparison. ACEL-Eval is designed to serve as a benchmark resource for Classical Chinese entity linking and as a reference for related computational studies of premodern Chinese historical texts.