Preprocessing and correction of mixture spectra have been an important issue with regard to the removal of undesired systematic variation due to variations in environmental, instrumental, or sample conditions. In this article, a new robust calibration modeling strategy is proposed on the basis of independent component analysis (ICA). It aims at separating the interference-subject parasitic subspace from the interference-immune common subspace among all considered cases. The common subspace is further divided into two orthogonal parts according to their relationship with quality: one is quality-irrelevant and the other is quality-informative, in which, only the second part is employed for quality prediction. Focusing on each subspace, it identifies distinct types of underlying source components underlying different spectra subspaces, analyzes their characteristics and roles, and accordingly models them for different applications, respectively. This approach provides a comprehensive insight into the inherent nature of interference-subject mixture spectra. Furthermore, several model statistics are defined to give quantitative indication on the effectiveness of the correction strategy. The feasibility and performance of the proposed method are illustrated with data from laboratory experiments. (C) 2009 American Institute of Chemical Engineers AIChE J, 56: 196-206, 2010