Abstract: In view of the undesirable effect of the classification accuracy in the process of the particle size identification of the excess solder particles caused by poor particle size discrimination and the intersection of the particle size characteristic parameters, a cluster-based method for the particle size identification is proposed. Based on the signal analysis techniques in the time-domain and the frequency-domain, we select a group of characteristic parameters to construct the eigenvectors for representing the particle size. The Fisher ratio is used to quantify the distinguishing ability of each feature parameter for the particle size identification, and those feature parameters with a low contribution rate are ignored in constructing the final particle size feature identification model. The K-means clustering algorithm is used to learn from the unlabeled training samples of different particle size levels and to obtain the distribution of the input characteristic parameters for different levels of particles, to realize the size identification for the mixed particle size. Testing results show that the overall recognition accuracy rate reaches 81.8% in the case of a single or two redundant objects, which can meet the application requirement.