Seed-based cross-correlation analysis (sCCA) and independent component analysis have been widely employed to extract functional networks from the resting state functional magnetic resonance imaging data. However, the results of sCCA, in terms of both connectivity strength and network topology, can be sensitive to seed selection variations. ICA avoids the potential problems due to seed selection, but choosing which component(s) to represent the network of interest could be subjective and problematic. In this study, we proposed a seed-based iterative cross-correlation analysis (siCCA) method for resting state brain network analysis. The method was applied to extract default mode network (DMN) and stable task control network (STCN) in two independent datasets acquired from normal adults. Compared with the networks obtained by traditional sCCA and ICA, the resting state networks produced by siCCA were found to be highly stable and independent on seed selection. siCCA was used to analyze DMN in first-episode major depressive disorder (MDD) patients. It was found that, in the MDD patients, the volume of DMN negatively correlated with the patients' social disability screening schedule scores.