Abstract: Traditional reconfigurable intelligent surface (RIS)-assisted direction of arrival (DOA) estimation methods suffer from high computational complexity and exhibit limited performance under conditions of undersampling and low signal-to-noise ratio (SNR). To address this issue, this paper proposes a RIS-assisted DOA estimation method based on the Bernoulli-Gaussian (BG) prior. To fully exploit the structured characteristics of incoming waves in the spatial domain, the proposed method introduces the BG prior model for modeling and performs Bayesian inference within the variational expectation maximization (EM) framework. Meanwhile, the damped generalized approximate message passing (GAMP) algorithm is embedded to reduce computational complexity. Simulation results demonstrate that, compared with traditional methods, the proposed method achieves higher direction estimation accuracy under conditions such as spatial sparsity and block sparsity of incoming waves, and enables stable estimation with fewer measurements and lower SNRs, which fully verifies the feasibility, efficiency, and robustness of the proposed method.