The invention discloses an automatic lie detection method and system based on domain adversarial training, and the method comprises the steps: S1, multi-modal feature extraction: extracting text feature representation, audio feature representation and facial feature representation; S2, performing multi-modal feature fusion, and obtaining multi-modal feature representation by using an adaptive attention mechanism; S3, performing time sequence modeling, and capturing context information in the dialogue by using a bidirectional recurrent neural network to assist lie detection of the current sentence; S4, performing domain adversarial training, extracting lie feature representation irrelevant to speakers by using a domain adversarial network, and reducing the influence of speaker difference onautomatic lie detection performance; S5, predicting a lie level, and inputting to-be-tested data into the lie classifier subjected to domain adversarial training for predicting the lie level of the individual. The system comprises a multi-modal feature extraction module, a multi-modal feature fusion module, a time sequence modeling module, a domain adversarial training module and a lie level prediction module which are sequentially connected from top to bottom.