Abstract: In a fermentation process several variables, such as biomass concentration are conventionally determined by off-line laboratory analysis, i.e., the process control is unavailable to industrial production in time just because of time delay that often makes the analysis results inefficient. In this respect, however, soft sensing is a good solution. Based individually on neural network and LS-SVM (least square support vector machine), two on-line soft sensing models are designed to estimate the biomass concentration, i.e., the black-box model and hybrid model. The data for model training and verifying are both got from a real experiment process-the fermentation of Nosiheptide which is also used to evaluate the model performance. The results show that the way of soft sensing is good at estimating the biomass concentration, and what's more higher estimating accuracy of hybrid model can be obtained if adding prior knowledge to the model.