Polymerization products are indispensable for our daily life, and the relevant modeling process plays a vital role in improving product quality. However, the model identification of the related process is a difficult point in industry due multivariate, nonlinear and time-varying characteristics. As for the conventional offline subspace identification methods, the identification accuracy may be not satisfying. To handle such a problem, an enhanced on-line recursive subspace identification method is presented on the basis of principal component analysis and sliding window (RSIMPCA-SW) in this paper to obtain the state space model for polymerization. In the proposed on-line subspace identification approach, the initial L-factor is acquired by the LQ decomposition of the sampled historical data, firstly, and then it is updated recursively through the bona fide method after the new data have been handled by the sliding window rule. Subsequently, principal component analysis (PCA) is introduced to calculate the extended observation matrix, and finally the on-line model parameters are extracted. Compared with the traditional subspace schemes, smaller computation complexity and higher identification precision are anticipated in the proposed method. A case study on the modeling of the ethylene polymerization verifies the effectiveness of the developed approach, in which the related statistical indexes of the obtained identification model are better.