The present invention discloses an encryption flow analysis feature selection method based on feature efficiency. The method comprises the steps of: defining a feature efficiency calculation method F(f); calculating the feature efficiency of each feature on two classes of samples, selecting the maximum one-dimensional features of the feature efficiency, removing samples of the one-dimensional features except the value domain overlapping, recording the number of the removed samples, and repeating in this way until all the features are calculated; and selecting features according to the pre-determined number of the features or the threshold values of the feature efficiency. The encryption flow analysis feature selection method can effectively calculate the feature efficiency of each featureon the premise of giving the maximum feature set, and can screen the features according to the given feature efficiency threshold values or the specified number of times to facilitate improvement of the identification accuracy of the website fingerprint recognition technology and save the time and space cost consumed in the classification model construction process.