Direct class II composite restorations still represent a challenge, particularly when proximal limits extend below the CEJ. The aim of this in vitro study was to evaluate the influence of the type of adhesive and the delay between adhesive placement and composite insertion on restoration adaptation. Direct class II MOD box-shaped composite restorations (n=8 per group) were placed on intact human third molars, with proximal margins 1mm above or under CEJ. All cavities were filled with a horizontal layering technique, immediately after adhesive placement (IP) or after a 24h delay (DP). A filled three-component adhesive (OptiBond FL: OB) and a single-bottle, unfilled one (Prime & Bond 2.1: PB) were tested. Marginal adaptation was assessed before and after each phase of mechanical loading (250000 cycles at 50 N, 250000 cycles at 75 N and 500000 cycles at 100 N); internal adaptation was evaluated after test completion. Gold-plated resin replicas were observed in the SEM and restoration quality evaluated in percentages of continuity (C) at the margins and within the internal interface, after sample section. Adaptation to beveled enamel proved satisfactory in all groups. After loading, adaptation to gingival dentin degraded more in PB-IP (C=55.1%) than PB-DP (C=86.9%) or OB-DP (C=89%). More internal defects were observed in PB samples (IP: C=79.2% and DP: C=86.3%) compared to OB samples (IP: C=97.4% and DP: C=98.3%). The filled adhesive (OB) produced a better adaptation than the 'one-bottle' brand (PB), hypothetically by forming a stress-absorbing layer, limiting the development of adhesive failures. Postponing occlusal loading (such as the indirect approach) improved also restoration adaptation.