The invention discloses an optimization method for an RFID (Radio Frequency Identification Devices) tag antenna. The optimization method comprises the following steps: step (1), a model input node acquisition step; step (2), a neural network model building step; and step (3), an RFID tag antenna optimization step, respectively selecting a plurality of values for performance parameters of the RFID tag within a reasonable range, combining into multiple groups of input nodes, respectively inputting the multiple groups of input nodes into the neural network model, correspondingly calculating output, that is a predicted reading distance of a reader, selecting a maximum predicted reading distance from multiple groups of predicted reading distances, and producing the RFID tag by the performance parameters of the RFID tag within the input node corresponding to the maximum predicted reading distance. According to the optimization method for the RFID tag antenna provided by the invention, through a method of predicting a tag performance and the reading distance by neural network modeling, data can be acquired only through a small amount of experiments, and a parameter combination value that all conditions affect the reading distance can be predicted, which saves both time and labor and the accuracy is very high.