Empathy is crucial for effective clinical care but appears to decline during undergraduate medical training. Understanding the nature of this decline is necessary for addressing it. The Jefferson Scale of Empathy (JSE) is used to measure medical students' clinical empathy attitudes. One recent study described a 3-factor model of the JSE. This model was found in responses from matriculating medical students, but little is known about how the factor structure of the scale changes during clinical training. The Learning Environment Study is a longitudinal prospective study of two cohorts from 28 medical schools. At matriculation and at the end of each subsequent year, students self-reported clinical empathy attitudes using the JSE. Data from 4,797 students were randomly partitioned for exploratory (EFA) and confirmatory factor analyses using responses from preclinical and clinical years of medical school. Five models were compared for confirmatory factor analysis: two null models for control, the recent 3-factor model, and the two models resulting from the EFAs of preclinical and clinical year responses. Preclinical year responses yielded a 3-factor model similar to the recent 3-factor model. Clinical year responses yielded a 4-factor model ("feelings," "importance," "ease," and "metacognitive effort") suggesting changes in the structure of clinical empathy attitudes over time. Metacognitive effort showed the largest decline over time. The model is a better fit for both preclinical and clinical responses and may provide more insight into medical students' clinical empathy attitudes than other models. The emergence of metacognitive effort in the clinical years suggests empathy may become more nuanced for students after clinical exposure and may account for much of the observed decline in clinical empathy attitudes.