Purpose: To investigate the influence of three cavity designs on the marginal seal of large Class II cavities restored with low-shrinkage resin composite limited to the enamel. Materials and Methods: One hundred twenty (120) intact human molars were randomly divided into 12 groups, with three different cavity designs: 1. undermined enamel, 2. box-shaped, and 3. proximal bevel. The teeth were restored with 1. an extra-low shrinkage (ELS) composite free of diluent monomers, 2. microhybrid composite (Herculite XRV), 3. nanohybrid composite (Filtek Supreme XTE), and 4. silorane-based composite (Filtek Silorane). After artificial aging by thermocycling and storage in physiological saline, epoxy resin replicas were prepared. To determine the integrity of the restorations' approximal margins, two methods were sequentially employed: 1. replicas were made of the 120 specimens and examined using SEM, and 2. the same 120 specimens were immersed in AgNO3 solution, and the dye penetration depth was observed with a light microscope. Statistical analysis was performed using the Kruskal-Wallis and the Dunn-Bonferroni tests. Results: After bevel preparation, SEM observations showed that restorations did not exhibit a higher percentage of continuous margin (SEM-analysis; p > 0.05), but more leakage was found than with the other cavity designs (p 0.05). The lowest percentage of continuous margin was observed in ELS restorations (p 0.05). More fractured margins were observed in the undermined enamel cavity design groups (p 0.05). Conclusion: Bevel preparation failed to improve margin quality in large Class II composite restorations and is no longer recommended. However, undermined enamel should be removed to prevent enamel fractures. Keywords: cavity design, enamel margins, low-shrinkage resin composite, bevel, box-shaped cavity, marginal gap, enamel fracture, undermined enamel