Composite material samples were created by means of computer simulation to duplicate short-fiberreinforced metal-matrix composites (MMCs). Each sample contains a fairly large number of Voronoi grains and ellipsoidal short fibers, which orient and distribute in a random manner, to mimic composite microstructures for investigating the coherent interconnections of composite-structure weaknesses (CSWs) with local microstructure. It is supposed that the samples are subjected to coupled boundary traction due to mechanical loading and thermal cycling. A Kröner-Kneer structure-based model and Waldvogel-Rodin algorithm were used for numerical computations of the mesoscopic stress distribution in constituent grains. The computations are based on the grain-volume average of local fields. Polycrystal elastic/thermal properties and effective elastic/thermal properties of simulated MMC samples were predicted, respectively, in terms of micromechanics models, in favor of incorporating the influences of macroscopic material properties on the formation of CSWs. An analytically-numerically-based approach is proposed for analyzing peak mesoscopic stress and strain distributions in short fibers. Three crucial aspects constitute a kernel of the approach, i.e., (1) segmentation of short fibers, (2) establishment of the geometric relations of a short fiber to the surrounding grains, and (3) the local nature of micromechanics. The analytically-numerically-based approach takes into account the grain orientation, fiber orientation, grain geometry, fiber geometry, and macroscopic properties of simulated MMC samples. The Numerical Assessment of Computer-Imitated Weaknesses-MMCs (NACIW-MMCs) software program has been developed for performing simulation of the microstructure of short-fiber-reinforced MMCs and executing all involved numerical computations.