Abstract: ‍ ‍Expression is an important way of human emotional interaction. Neurophysiological studies show that micro-expressions （MEs） are not controlled by subjective consciousness and reflect people's real emotions. Unlike macro-expressions， micro-expressions are often accompanied by asymmetric facial movements with complex movement patterns. However， due to the small amplitude of micro-expression， it is difficult to be directly observed by humans. The ME movement law has not been deeply analyzed. The reliability and interpretability of ME recognition algorithms are highly required in public security. Therefore， this paper aimed to study the micro-expression movement law analysis method. The main work of this paper is as follows. We studied the unsupervised clustering method for ME features through self-organizing maps （SOM） network to obtain micro-expression movement patterns. The micro-expression movement distance （DME） was defined， which measured the difference between two micro-expression features in terms of movement in the fourteen regions of interest on the face and served as a basis for adjusting the weights of the SOM network. In the experimental part， we analyzed micro-expression samples from CASMEII， SAMM， SMIC and MMEW datasets and summarized micro-expression movement law based on the learning results of the SOM network. This law can effectively guide the feature extraction of micro-expression recognition algorithm and improve the reliability.