Abstract: A Multi Neural Network method is proposed for soft sensing in a complex non-linear biochemical process. Low dimensional sample data are achieved by way of original data compression through principal component analysis, and a kind of modified sorting indices are used for these data to accomplish data hierarchy of the biochemical process model. Then, a soft sensing model is developed using multi neural network to fit the different hierarchical property of the process. The method has been applied to a glutamic acid fermentation process, and the testing results showed its effectiveness.