A number of rule-of-mixture micromechanical models have been successfully used to predict mechanical properties of short fiber composites. However, in order to obtain accurate predictions, detailed description of the internal structure of the material is required. This information is often obtained from optical microscopy of polished cross-sections of composite. This approach gives accurate yet local results, but a rather large number of optical images have to be processed to achieve a representative description of the morphology of the material. In this work a fully automatic algorithm for estimating the length-diameter distribution of solid or hollow fibers, utilizing three-dimensional X-ray tomographic images, is presented. The method is based on a granulometric approach for fiber length distribution measurement, combined with a novel algorithm that relates cross-sectional fiber properties to fiber length. The work opens up a possibility to assess multivariate distributions of fiber length and diameter, cross-sectional area or other microstructural fiber properties. As an example, the description of microstructure of different composites with natural fibers is presented, along with verification of the results.