A method is presented for estimating direction of arrival (DOA) of multiple spatio-temporal sources in the domain, which is based on independent component analysis (ICA). Firstly, original time-domain data is dimensionally reduced by using some classical second-order techniques for estimating number of independent sources (NIS), such as singular value decomposition (SVD) and eigenvalue decomposition (EVD). From every reduced data segment, steering vectors of the mixing system from sources to sensors are directly identified by some instantaneous ICA algorithms. Furthermore, DOAs of independent sources are captured by using a whole estimating strategy. Consequently, all time-domain data segments contribute to a final DOA estimation set, in which every source direction is shown as a direction cluster and/or local maximum. Experimental results indicate potential applicability of the proposed ICA based methods.