The usual characterization of particulate ore material with image analysis includes modal analysis and liberation analysis. Both methods include stereological corrections based on intercept length distributions within each phase of interest. The first principle of stereology relies on a simple assumption of uniform random sectioning, whereas the liberation models require all particles to have similar textural properties. However, for the sake of adequate prediction of ore behavior in industrial processes it is becoming more and more important to be able to describe particle populations in terms of texture classes. Among the texture analysis techniques, this paper focuses on mean intercept length (MIL). It brings two important improvements: the first one is to improve grain boundary imaging by using images of bireflectant minerals under plane polarized illumination, the second one is to consider intercept analysis on grey level images instead of binary ones. The method is illustrated on a selection of critical hematite textures found within banded iron formations. (C) 2007 Elsevier B.V. All rights reserved.