The invention provides a micro-expression recognition method and device. The method comprises the steps that a micro-expression test set is acquired; local binary pattern LBP-TOP feature vectors in three orthogonal planes are extracted for each micro-expression picture in the test set, and the obtained feature vectors are used as predictors of a group sparse least square regression (GSLSR) model; and tags corresponding to the GSLSR model in the test set of the micro-expression pictures are acquired according to the obtained predictors of the GSLSR model for each micro-expression picture through the relation between sequence tag vectors and sequence feature vectors of the micro-expression pictures. In this way, the feedback quantity of a receiving terminal is lowered, and information interaction between transmitting terminals is omitted.