This paper presents a new framework towards control of cross-directional profiles in sheet-forming processes. The proposed method uses adaptive principal component analysis to identify the significant disturbance modes from on-line data. To resolve the potential problems associated with large dimensional and ill-conditioned systems a low order subspace is constructed for controller design purposes. Certain projection and reduction techniques are given to systematically arrive at this design subspace. Simulations illustrate that the resulting controller is easy to design and able to isolate and reject the significant disturbance modes.