Existing models for built-in residual stresses in composite materials are reviewed and discussed. In particular, the thermal longitudinal stress present in the fiber prior to a single-fiber fragmentation experiment is studied using various model composite data. It is found that this stress is typically compressive in nature and that, quantitatively, it depends on the fiber content, the degree of undercooling, and the thermoelastic constants of the fiber and the matrix. In the case of single-fiber composites (or microcomposites), the thermal longitudinal stress present in the fiber is high enough to either induce fiber sinewave buckling (such as in E-glass/epoxy), or extensive fiber fragmentation (such as in graphite HM/polypropylene) that may then be used to measure the dependence of compressive fiber strength upon length. This has to be accounted for in quantitative models that calculate interfacial adhesion parameters using single-fiber tests, such as the fragmentation test or the microbond test. Implications for high fiber content composites (or macrocomposites) are discussed.