To overcome the drawbacks of manual quality inspection in battery industry, an online vision system is designed for battery screen print. Defect detection technique is based on the joint method of multi-level block matching and K nearest neighbor (KNN) algorithm. Firstly, execute preprocessing to origin images in segmentation, tilt correction and region cutting; Then create block templates on print area and train the corresponding models for active shape model (ASM) and KNN methods; Finally, coarse and accurate block matchings are applied to extract print defects in subsequent stages. In this period, KNN uses shape features of region components to recheck each target block. In addition, we adopt dynamic model updating mechanism to enhance system adaptability of condition changing. The joint method has two advantages: fault detection caused by print distortion is obviously reduced; accurate defect localization is also assured. Meanwhile, system hardware and software are also developed and calibrated to support detection method. Performance comparison, recognition rate and time efficiency are validated in experiment stage. It can be concluded that the proposed method has superior performances in both simulations and industrial application. © 2021 Elsevier GmbH