Abstract: Thick shield and small detector problem often occurs in reactor shielding calculation. It is difficult to solve it by the conventional Monte Carlo method. A variance reduction method called small detector auto-importance sampling (SDAIS) method was proposed for the thick shield and small detector problem. The SDAIS method is the improved method of the auto-importance sampling (AIS) method in the case of small detector problem. The virtual particle Russian roulette method was used based on the detector position instead of the weight. The SDAIS method was implemented in the self-developed Monte Carlo program MCShield. The NUREG/CR-6115 PWR benchmark was used to verify the correctness and computational efficiency of the method. The results show that compared with the AIS method and the traditional importance method, the calculation efficiency of SDAIS method has an improvement of 1-2 orders of magnitude. SDAIS method can effectively solve the problem of thick shield and small detector.