The two-stage grinding size in dressing work is a key factor affecting the iron concentrate grade and recovery rate. Aiming at the problem that the current dressing work cannot detect the grinding size in real time, combined with the two-stage grinding process, the method based on particle swarm optimization (PSO) is used to optimize the T-S fuzzy neural network, and a soft measurement model of two-stage grinding size is established; while the actual production data of the dressing work is collected, the model comparison tests is conducted. The results show that the model hit rate of the T-S fuzzy neural network optimized by PSO is 94%, the average relative error is 0.005 4, and the model performance is significantly better than the T-S fuzzy neural network model and the RBF neural network model; the optimization model can effectively solve the fuzzy problem between the particle size and variables of the two-stage grinding, and the prediction accuracy is high, which meets the requirements of the dressing work for the real-time detection of the size of the two-stage grinding. The research results can provide a new strategy for modeling the soft measurement of the size of the two-stage grinding.