One important application of natural language processing (NLP) is the recognition of emotions in text. Most current emotion analyzers use a set of linguistic features such as emotion lexicons, n-grams, word embeddings, and emoticons. This study proposes a new strategy to perform emotion recognition, which is based on the homologous structure of emotions and narratives. It is argued that emotions and narratives share both a goal-based structure and an evaluation structure. The new strategy was tested in an empirical study with 117 participants who recounted two narratives about their past emotional experiences, including one positive and one negative episode. Immediately after narrating each episode, the participants reported their current affective state using the Affect Grid. The goal-based structure and evaluation structure of the narratives were analyzed with a hybrid method. First, a linguistic analysis of the texts was carried out, including tokenization, lemmatization, part-of-speech tagging, and morphological analysis. Second, an extensive set of rule-based algorithms was used to analyze the goal-based structure of, and evaluations in, the narratives. Third, the output was fed into machine learning classifiers of narrative structural features that previously proved to be effective predictors of the narrator’s current affective state. This hybrid procedure yielded a high average F1 score (0.72). The results are discussed in terms of the benefits of employing narrative structure analysis in NLP-based emotion recognition.