Abstract: In order to improve the quality of chlortetracycline fermentation, a multiphase quality prediction system based upon dynamically dividing phase of the culture was proposed. By clustering the PCA loading matrices of history dataset to obtain the clustering center of each phase, the fermentation phase can be divided online through calculating the Euclidean between current loading matrix and these ones; while quality prediction was modeled for each phase of the chlortetracycline fermentation with relevance vector machine algorithm. During chlortetracycline fermentation batch, a corresponding control strategy was implemented on the fermentation production in terms of the information of phase and quality value. The validity and reliability of the improvement of product quality and product yield of the proposed method was illustrated by applying it to the real chlortetracycline fermentation.