We propose a technique, called source-space-ICA to provide spatiotemporal reconstruction of brain sources. First, the weight-vector-normalized minimum variance beamformer is applied to reconstruct the electrical activity of a 3D scanning grid which covers the whole brain. Second, principal component analysis is used to reduce the dimension of the reconstructed signal matrix of the source-space, then independent component analysis (ICA) is applied on the resulting matrix to identify multiple signal sources in the source-space. Third, the demixing weight vectors obtained by ICA for the identified independent components are projected back into the SS to obtain tomographic maps of the sources. Besides localization, the proposed source-space-ICA approach reconstructs the time-course of each source in a single time-series without requiring prior knowledge of location, orientation, and number of sources for a given segment of EEG/MEG. Simulated EEG was used to evaluate the source-space-ICA. The results show that the source-space-ICA approach is able to separate and localize multiple weak sources and is robust to interference from other sources as it identifies the sources based on their statistical independence.