Testing the influence of various restorative materials (conventional or bulkfill composites) and filling techniques (single- or multi-layered techniques) on the in vitro marginal adaptation of large class II direct composites with supra and sub-gingival margins subjected to thermomechanical loading (TML). Methods: A total of 40 prepared teeth were attributed randomly to five experimental groups. In Group 1, restorations were made of multi-layered high-viscosity conventional composite (Tetric EvoCeram); in Groups 2 and 3, restorations were made of a high viscosity bulkfill composite (Tetric EvoCeram Bulk Fill) applied in one (Group 2) or three layers (Group 3); in Groups 4 and 5, restorations were made of a flowable bulkfill composite (SDR flow) applied in one (Group 4) or two layers (Group 5), covered with a layer of high-viscosity conventional composite (Ceram-X Universal). A single adhesive (OptiBond FL) was used in all groups. All specimens were submitted to a staged TML comprising three phases (2 × 500,000 and 1,000,000 cycles) at 50 N with 3350 thermal cycles (5 to 55 ◦C) every 500,000 cycles. The tooth–restoration interfaces on proximal surfaces were evaluated quantitatively by scanning electron microscopy, before and after each TML phase, hence at three timepoints (T0, T1, T2 and T3). The following segments were considered for evaluation: proximal, vertical enamel margins (assessed individually on both restoration sides), cervical enamel (restoration side above CEJ) and cervical dentin margin (restoration side below the CEJ). Results: TML induced a significant reduction in continuous adaptation at both enamel and dentin margins in all groups, with percentages of continuous margins ranging from 75.2 to 91.8% at T0, and decreasing to values ranging from 21.3 to 73.9% at T3. Both composite systems and layering protocols had a significant influence on marginal adaptation of the restorations, with statistical associations depending on the restoration group and the timepoint considered. Defective margins in enamel were all of a cohesive nature with micro-fractures, while in dentin, interfacial gaps were the main defect observed. Conclusions/Clinical significance: The present results highlighted significant degradation of marginal adaption after long-term in vitro fatigue test using materials even with high-viscosity conventional resin composites, applied with a proper layering approach in medium–large sub-gingival cavities. While no significant differences were observed at the dentin cervical margins, there was a tendency for better adaptation at the enamel margin when using a higher modulus material with a multi-layered technique