In this paper, we present an improved approach to the �breaking the F5 algorithm�. The key idea is to search for an �optimal� value of β (the probability that a non-zero AC coefficient will be modified) for the image under consideration. Rather than averaging the values of β over 64 shifting steps worked on an image, an optimal β is determined that corresponds to the shift having minimal distance E from the double compression removal step. Numerical experiments were carried out to validate the proposed improved algorithm and compare it against the original one using two datasets of test images. The first dataset involves 432 images created by manipulating 12 images for various image parameters, and the second dataset uses reference benchmark 20 grayscale images used by the original algorithm,. Test results show that the proposed algorithm yields less average absolute error values for β and secret message length than the original algorithm; this is particularly valid for stego images with actual β values larger than 0.05. We suggest that the original algorithm may be used as a classifier, since it shows a good detection performance of both clean and stego test images; whereas, the proposed algorithm may be used as an improved estimator for the true message length for those images that have been classified by original algorithm as stego images.