Complete removal of failed posterior resin composite fillings is time-expensive and involves the risk of removing sound tooth substance. In any case of failure within the composite material, intra-oral repair would be favourable. This in vitro study was conducted to examine the effect of different preparation and bonding modes on integrity of repaired restorations. Forty-eight direct Class II resin composite restorations (Syntac Classic/Tetric Ceram) were placed in extracted human third molars. The specimens were stored for 365 days and then replaced partially by removing the central part of the proximal box with all margins located in composite. The partial repairs were placed with the same materials after pre-treatment with a silicon carbide bur and Syntac Classic (n = 8). Preparation modes have been (i) box-only parallel, (ii) box-only with undercuts, (iii) box with undercuts and additional occlusal retention. The repairs were performed either with (i) Tetric Ceram or with (ii) Tetric Flow as thin lining covered with Tetric Ceram. Before and after thermomechanical loading (100 000 cycles of 50 N and 2500 cycles between +5 and +55 degrees C in an artificial mouth), marginal quality between aged and freshly applied composite was evaluated by SEM at x200 magnification and microleakage was observed by light microscopy at x50 magnification. Box-only preparations exhibited a better fatigue resistance than preparations with additional occlusal retention. This observation was independent from the presence of undercuts (P < 0.05). An intermediary layer of flowable resin composite tended to result in better marginal quality, however, not being statistically significant.