AbstractParticle granulometry plays an important role in the engineering behavior of many sands. However, the evaluation of particle shape and size has historically been a tedious and labor-intensive process. The recent availability of dynamic image analysis (DIA) makes it possible to evaluate many particle shape and size parameters, quickly and conveniently. These shape parameters include sphericity, roundness, aspect ratio, circularity, and convexity; while size descriptors include the diameter of a circle of equal projection area (EQPC), a variety of Feret diameters, as well as inscribed and circumscribed circle diameters. The terms roundness and sphericity are commonly used to describe how close a particle resembles a sphere, with many definitions in common use. However, it is not immediately evident how these roundness descriptors correlate. The correlation of nine shape and six size descriptors was investigated for six sands that reflect the breadth of particle shapes and sizes that may be encountered. The analysis was based on 1,000 images of each sand obtained using two-dimensional DIA apparatus. The study demonstrates that there is no correlation between size and shape parameters, and that shape descriptors can be reduced to four independent shape parameters representing the granulometry of sand at different scales. The use of size and shape descriptors for classification of sand was explored using six machine learning algorithms including support vector machines (SVMs), random forest, decision tree, bagging tree, k-nearest neighbors (KNN), and bagging KNN. Classification accuracies of 77% and 66% were achieved using size and shape features, respectively. The mean accuracy improved to 87% when combining both size and shape descriptors using bagging KNN and random forest classifiers. The analysis also revealed an important hierarchy of size and shape features employed, with EQPC and Wadell’s roundness alone classifying sands with 70% accuracy.