A method and device for training a classification model, relating to the technical field of information. The method for training a classification model comprises: obtaining a sample characteristic vector of each of multiple users according to at least one characteristic set of each user (201); determining a gender identifier of the user to which each sample characteristic vector belongs (202); and training according to each sample characteristic vector and the gender identifier corresponding to each sample characteristic vector to obtain a gender classification model (203). At least one characteristic set of each user is determined according to at least one sample SMS message of each user, so as to obtain a sample characteristic vector of each user; a gender identifier of the user to which each sample characteristic vector belongs is determined; training is performed according to multiple sample characteristic vectors and the gender identifier corresponding to each sample characteristic vector to obtain a gen
