The background of the present study is the rolling contact fatigue (RCF) in a brittle polymer disk. The disk has been tested on a two disk machine, under controlled normal and tangential loads, with no global slip. After several million cycles and under different operating conditions, it has been observed that (1) the tangential load highly influences the RCF phenomenon, (2) a network of regularly spaced cracks appears, and (3) in the driving position, the RCF phenomenon develops faster. To explain these observations, a numerical model based on the finite element method (FEM) has been built: the cracks have been quite simply modeled, stick-slip has been chosen as the friction model, and the disk-on-disk contact has been replaced by a disk-on-plane contact. To study the influence of some of the operating conditions, the design of experiments (DOE) techniques has been used. The statistical postprocessing associated to DOE has confirmed the experimental observations with a good reliability. In addition, with some mechanical considerations, scenarios of what experimentally happens are proposed. The association FEM/DOE is an original and efficient way to explain phenomena in the field of RCF: the accuracy of the FEM coupled with DOE statistical treatments make it possible to have a good predictability despite some uncontrolled parameters.