Starch granules that accumulate in the plastids of plants vary in size, shape, phosphate, or protein content according to their botanical origin. Depending on their size, the applications in food and nonfood industries differ. Being able to master starch granule size for a specific plant, without alteration of other characteristics (phosphate content, protein content, etc.), is challenging. The development of a simple and effective screening method to determine the size and shape of starch granules in a plant population is therefore of prime interest. In this study, we propose a new method, NegFluo, that combines negative confocal autofluorescence imaging in leaf and machine learning (ML)-based image analysis. It provides a fast, automated, and easy-to-use pipeline for both in situ starch granule imaging and its morphological analysis. NegFluo was applied to Arabidopsis leaves of wild-type and ss4 mutant plants. We validated its accuracy by comparing morphological quantifications using NegFluo and state-of-the-art methods relying either on starch granule purification or on preparation-intensive electron microscopy combined with manual image analysis. NegFluo thus opens the way to fast in situ analysis of starch granules.