The invention relates to the technical field of RFID, in particular to the field of RFID multi-tag distribution configuration, and belongs to the technical field of detection; and especially Flood-Fill and an SVM neural network are introduced for predicting RFID tag distribution to obtain optimal configuration. The invention provides a Flood-Fill and SVM-based RFID multi-tag three-dimensional optimal distribution detection method. RFID tags on the surface of cartons are subjected to position extraction through the Flood-Fill, and reading distances corresponding to RFID multi-tag positions are trained through the SVM neural network, so that the RFID multi-tag positions under the specific reading distances are predicted, the RFID tag distribution with optimal reading performance is found, and the influence of an actual working environment on the reading performance is reduced from the perspective of RFID tag distribution optimization configuration. The method can effectively improve the reading performance of the RFID tags and has important theoretical and application values for RFID technology development.