A machine-vision based defect detection method for packaging bags is proposed. Considering the defect detection of ice-lolly bags as an example, five kinds of eigenvalues of length, width, area, filling degree, and location relationship between the monitoring frame and the internal target region are extracted. After defect detection and classification, the following four types of defects are outputted: continuous bags, dimension errors, foreign matters on packages, and motion of packaging layout. The experimental results demonstrate that the recognition success rate of the proposed algorithm can reach 98.75%, which meets the requirements of high speed, high precision, and real time in the production process. The algorithm has been applied to an actual production line and has achieved good results.