Analyzing broad and strongly overlapping far-field scalp projections of underlying spatially distinct locally-synchronous cortical field activities has long posed a challenge for cognitive neuroscience researchers. Even the accurate forward solution was obtained, the source localization problem is still highly underdetermined if multiple sources contribute to the EEG recordings. To attack the indeterminate nature of the EEG source analysis problem, some constraints were usually applied. The linear regulation‐based methods (eg. minimum-norm estimators, low resolution tomography, LORETA, etc.) drive solutions to smoothness, whereas parametric dipole fitting and Sparse Bayesian Learning (SBL) algorithms drive source to an opposite direction, sparsity. Here, we propose a new source imaging method whose goal is to obtain more physiologically realistic solutions to the EEG inverse problem by combining a priori knowledge about nature and structure of brain sources, including spatiotemporal independence, sparsity, spatial compactness and local smoothness. The performance of the proposed method is evaluated qualitatively by using experimental and simulated EEG data. (Coauthored with Chii-Shyang Kuo, Cheng Cao, Michelle Liou and Scott Makeig) Keywords: EEG, source localization, spatiotemporal independent component analysis.