OBJECTIVES:   Multiple sclerosis (MS) is a disease with lifelong impact, making the cost-effectiveness (CE) of its treatments particularly sensitive to assumptions embedded in model designs. Traditional sensitivity analysis (SA) can test many assumptions, but it is not designed to investigate sensitivity to structural assumptions. The aim of this study was to compare a Markov cohort model (MM) and a discrete-event simulation (DES) model of MS that were based on common clinical data but developed independently to understand the impact of their structural differences on model predictions. METHODS:   A similar population was simulated in the MM and the DES model; aggregated cost and utility estimates were compared over varying time horizons. The average expanded disability status scale (EDSS) and the distribution of EDSS were also compared over time to study the dynamics of disease progression and treatment effects. RESULTS:  The two modeling approaches led to different natural history behavior over longer time horizons, even after short-term behaviors were well-aligned, with the DES model predicting slightly fewer life-years (25.9 vs. 26.2 in the MM) but more quality-adjusted life-years (9.6 vs. 8.1 in the MM). These differences reflect slower progression of EDSS in the DES model, particularly to higher EDSS states. When disease history (including a baseline EDSS term) was excluded from the DES model, the natural history simulations of the two models agreed more closely. CONCLUSIONS:   Structural SA can help quantify the impact of key modeling decisions.  In this study, a comparison of an MM and a DES model showed that natural history predictions diverge over long time horizons, in part due to the consideration of disease history in the DES model. A better understanding of the differences between the two model designs helps ensure interpretation of the model results while taking into consideration the assumptions embedded in those designs.