The invention relates to a method (100) implemented in an apparatus for identifying a hand gesture (1) of a user (2) which includes locating (101) a hand in a three-dimensional space (3) and obtaining (102) at least one two-dimensional projection (4) from a cloud of three-dimensional points (5) that represent said hand. For each two-dimensional projection obtained the following steps are performed: scaling (103) the two-dimensional projection; extracting (104) from the scaled two-dimensional projection a vector of features (v); normalising (105) each feature and identifying (106) a candidate hand gesture from the normalised vector of features by means of a previously trained classifier.