Agglomeration is one of the most challenging problems due to overheating of the particles in fluidized bed reactors (FBRs). Therefore, it is an urgent task to develop a reliable and sensitive method, which can help accurately detect the agglomeration in an early stage. In this study, acoustic emission early agglomeration recognition system (AE-EARS) has been put forward for fault detection. Based on acoustic emission signals, the attractor comparison method was developed for prewarning the agglomeration in lab-scale and pilot-scale apparatus. The results concluded from this study demonstrated that the statistical characteristic S acts more sensitively to small agglomeration when compared with the malfunction coefficients C-D2 and C-K2, and other traditional measurement techniques (such as gamma ray, temperature, and pressure difference). Besides, model optimization based on AE-EARS can help to select the criterion and improve the rate of false alarm. The analysis methods based on AE-EARS can warn the agglomeration earlier, faster, and more accurately. Especially the S value based on the attractor comparison, can be used as an indicator for "early recognition", which enjoys a broad prospect in industrial application.