The invention discloses an ADHD classification diagnosis method based on brain connection dual subspace learning. The method comprises the steps: making different types of hypothesis annotations for test data, performing the typical brain connection feature selection process of the test data subjected to hypothesis labeling and the training data, and obtaining typical brain connection features ofthe test data corresponding to the training data under different hypotheses; performing dual-subspace learning on the typical brain connection features of the test data corresponding to the training data under different hypotheses, and calculating the energy sum of the typical brain connection features of the training data on the dual-subspace under the corresponding hypotheses; comparing the energy sum of the typical brain connection characteristics of the training data in the dual subspace under different hypotheses of the test data, selecting the hypothesis with the large energy sum as thereal hypothesis of the test data, determining that the individual to which the test data belongs is an ADHD disease or a healthy individual, and finally completing the ADHD disease diagnosis and classification process. The classification accuracy of the method is about 90%, and the method is superior to various existing classification diagnosis methods.