We present a methodology for characterizing and reconstructing in-plane weave variability in textile composites. Surface topography of a partially processed C-fiber/SiC matrix composite panel was measured using digital image correlation. The centroids of tow segments that appear periodically on the fabric surface were located by image analysis and used as fiducial markers. Stochastic deviations of the fiducial markers from the ideal periodic weave structure indicate geometrical variance. Fourier analysis shows that spatial wavelengths of the deviations range from the size of one unit cell to the dimensions of the entire panel. Long-range deviations are attributed principally to fabric deformation after manufacture, during handling. Short-range fluctuations, extracted by computing spatial derivatives of the positions of the fiducial markers, are attributed to variations in tow packing density that arises during weaving. A simple set of statistics for these deviations is presented and its use in generating stochastic virtual specimens is demonstrated.