A method for automatically generating an image description includes following steps of: obtaining a first image data; analyzing a text file corresponding to the first image data; calculating occurrences and distribution ratios of terms, each of which contains a target word, from the text file, so as to obtain a plurality of the terms having the distribution ratios greater than a threshold; comparing the distribution ratios of the terms so as to find out at least one key term; finding out one of a plurality of lexical chains containing the key term with a greatest distribution ratio so as to generate a narrative language chain; and setting the narrative language chain as a description of the first image data. A system for automatically generating an image description and a non-volatile computer-readable storage media are also disclosed.