Abstract: Searching of lost radioactive sources rapidly and accurately is the core technology for radioactive source locating by aerial measurement. In order to solve the problem of radioactive source location in shielded condition, an adaptive radioactive source location algorithm based on Markov chain Monte Carlo (MCMC) method was proposed. Firstly, the initial values of radioactive source parameters were randomly given, and the physical model of the lost radioactive source parameter estimation was established by parameterizing the shielding attenuation coefficient. Then, based on the Metropolis-Hastings sampling method, the parameters of the radioactive source were sampled to generate Markov chain, and the appropriate sample set was selected to calculate the posterior distribution of the parameters of the radioactive source. Finally, the convergence state of the Markov chain was judged in real time, and the length of the sample set was adaptive, which can greatly shorten the time while ensuring the localization accuracy. To verify the feasibility and effectiveness of this method, based on numerical simulation and field experiment, the radiation field without shielding and the marble shielding radiation field with unknown attenuation coefficient were designed respectively. The original MCMC method, the MCMC method of attenuation coefficient parameterized and the MCMC method of adaptive sample set with convergence judgment were used to locate the radioactive sources respectively. The simulation and experimental results show that the proposed algorithm can locate radioactive sources in the environment with unknown attenuation coefficient shielding. Within the 60 m×60 m simulation search range, the localization accuracy is better than 1 m. Within the 5.4 m×2.4 m field experimental search range, the localization accuracy is better than 0.1 m, and the accuracy of localization is at the same level of magnitude as that in the environment without shielding. Due to the presence of shielding, the number of unknown parameters is increased, and the running time of the program is increased compared with that without shielding. However, by judging the convergence state in real time, it can greatly shorten the time while ensuring the accuracy of localization. Under experimental conditions, there is little difference in accuracy before and after realtime judgment of convergence state, but the time of the latter is about 10% of the former. In addition, the search environment for real radioactive sources is more complex, and in the actual operation process, the influence of irrelevant factors on measurement should be minimized as much as possible to improve positioning accuracy.