Baeillus cereus DM423 is a new replacement of antibody in stock breeding. In this paper, the biomass of Baeillus cereus DM423 during batch cultivation was measured by the soft sensor of time-delayed neural network, which was constructed with the topology of 11-20-1, the input variables was delays of pH, temperature, dissolved oxygen, glucose concentration at t–1 and t–2 and delays of biomass concentration at t–1, t–2 and t–3, and the output variable was biomass concentration at present time. The result showed that the constructed network had good generalization with the mean square error of 0.15 ?10–3 for the testing samples and good robustness and prediction ability.