Abstract: In order to solve the problem of the existence of gross errors in data samples for soft sensing modeling,the complexity of the dynamic recursive fuzzy neural network's structure,and the difficulty in determining the massive parameters, a soft sensor based on immune genetic algorithm and dynamic recursive fuzzy neural network is proposed.Similarities between samples are analyzed by the way of computing Mahalanobis distance,the gross errors in data sample are removed to increase the computing speed.In addition,subtractive clustering is applied to determining the number of fuzzy rules in order to simplify the network structure,and at the same time an immune genetic algorithm is introduced to optimize the model parameters to enhance and its precision and generalization ability.The method is applied to biomass concentration soft measurement in the lysine fermentation process.The simulation example shows that the model has high prediction precision and good generalization ability,and it satisfies the need of spot measurement.