Abstract. Particulate reinforced metal matrix composites (PRMMCs) are characterized by their stochastic and irregular microstructure. The strength of the material has been noticed to have a large scatter and great dependence on the underlying composite structure. In order to understand how size and distribution of the reinforcement particles contribute to the effective material behaviors, we elaborated in present study a numerical approach which incorporates homogenization technique, shakedown analysis, and statistical learning. To demonstrate this approach, a typical PRMMC material, WC -20 wt.% Co, was taken as example and numerous representative volume element (RVE) samples of this material were built based on both real and artificial microscope images. Multiple effective behaviors, including the ultimate strength and endurance limit, were evaluated on each RVE sample using the direct method. In order to understand how multiple structural factors jointly affect the overall composite strength, the predicted strength and other selected features extracted from the samples are submitted to few statistical models such as logistic regression and artificial neural network (ANN) to establish predictive models. On the basis of these models we discussed how few structural factors, which have been identified to have nontrivial impact, contribute to the endurance limit of the selected PRMMC material.