Fragmented rock size distribution is one of the most important parameters in open pit blasting that can affect the mining and the mineral processing efficiency. For evaluating fragmentation by blasting, digital image analysis is a fast and reliable indirect technique. In this study, based on the neural network and features extraction methods, an algorithm was proposed to determine the size distribution of fragmented rocks. For this purpose, 226 images of fragmented rocks from various blasts, carried out at Gole-Gohar iron ore mine, Iran were used to prepare a dataset. To extract visual features of these images, Fourier transforms, Gabor, wavelet methods and their combinations were used and features extracted considered as the input vectors of neural network. Also, for these images, using the manual mode of Split-Desktop software, F-10 to F-100 were determined (as the target data of neural network). Then, the results of features extraction methods for 26 test images were compared with the results of auto mode of Split-Desktop. The results obtained showed improvements in the estimation of fragmented rock size distribution using Fourier transform, Gabor and Fourier-wavelet methods with the value of 67%, 57%, and 48%, respectively. Also, the estimation of fragmented rock size distribution has higher MRE in fine to medium particles (F-10-F-50). Moreover, for F-10-F-50 the most improvements with the values of 52%, 40%, and 32%, are corresponding to Fourier transform, Gabor and Fourier-Gabor methods, respectively. Also, all of the suggested features extraction methods for estimating uniformity coefficient give better results than auto mode of Split-Desktop. (C) 2018 Elsevier B.V. All rights reserved.