This presentation examines the initial results of Visual Constituent Analysis (hereafter ‘VCA’); a novel, vision-based approach to creating multimodal corpora from text and imag-es. As opposed to classic approaches analysing text-image relation, where text functions as context to the visual stimuli, or vice versa, VCA presents images as a series of individual se-miotic constituents, which can then be shown in-line with the text. The approach comprises a new method of corpus construction that efficiently processes large amounts of image data, allowing for the use of traditional text-based corpus methods.