The occurrence of agglomeration in fluidized-bed production of polyolefin can have negative impacts on the efficiency of the reactor operation and may lead to defluidization and unscheduled shutdown of the plant. A novel method by using acoustic emission (AE) sensors was developed to detect different types of agglomeration in fluidized bed. Chaos analysis was discussed, and coefficients of malfunction were defined to recognize agglomeration. AE signals we divided into micro-, meso-, and macroscales by wavelet transform, and on the basis of the energy of mesoscale fraction analysis of AE signals, a prediction model was developed to predict the size of moving chunks in the bed. The application effect of AE method was illustrated with experiments carried out both in laboratory and the plant.