Sentiment analysis has gained widespread adoption in many fields, but not--until now--in literary studies. Scholars have lacked a robust methodology that adapts the tool to the skills and questions central to literary scholars. Also lacking has been quantitative data to help the scholar choose between the many models. Which model is best for which narrative, and why? By comparing over three dozen models, including the latest Deep Learning AI, the author details how to choose the correct model--or set of models--depending on the unique affective fingerprint of a narrative. The author also demonstrates how to combine a clustered close reading of textual cruxes in order to interpret a narrative. By analyzing a diverse and cross-cultural range of texts in a series of case studies, the Element highlights new insights into the many shapes of stories.