Aiming at resistivity inversion and uncertainty quantification in electrical impedance tomography （EIT）， an uncertainty analysis method is proposed based on Bayesian theory. Firstly， the Back Propagation （BP） neural network model is used as a substitute model for the forward problem， the results with high calculation accuracy are obtained， and the calculation efficiency is greatly improved. Then， the Differential Evolution Adaptive Metropolis （DREAM_ZS） sampling algorithm based on Bayesian theory is used for the resistivity reconstruction， and different excitation modes and prior distributions are compared and analyzed. The inversion results of the four-layer concentric circle model simulating the head show that the DREAM_ZS sampling algorithm can accurately identify the four parameters， and the inversion effect of the relative excitation mode is the best. The uncertainty of the four parameters is different. The scalp resistivity has the minimum uncertainty and the strongest sensitivity， and then the skull， the brain， and the cerebrospinal fluid show the maximum uncertainty. Furthermore， the circular model with high-dimensional parameters is simulated， and the relative excitation mode is adopted. DREAM_ZS sampling algorithm can accurately invert the parameters of the two-dimensional circular model. When the prior distribution of the parameters is normal distribution， compared with the uniform distribution， the uncertainty of the inversion result is less， and the recognition effect of the algorithm is better.