Over the past decade, meta-analyses have been conducted to evaluate the impact of embodied pedagogical agents on learning outcomes. Most review studies have evaluated learning outcomes from the perspective of testing labels such as transfer, retention, recognition, free recall, and various other classifications. This is problematic, because studies can be excluded due to the labeling of the assessment, even though the testing format and test taking strategy are the same. This meta-analysis assesses how embodied pedagogical agents impact learning when measured from the testing format and test taking strategy perspective. The new approach, along with new moderating variables to evaluate agent design and content information, supports previous findings that embodied pedagogical agents significantly increase learning; but there are limitations depending on the agent's design and the type of information being learned. This meta-analysis advances the field by exploring new ways to think about design elements associated with embodied pedagogical agents.