F. Souza, MultiMechanicsFiber reinforced composites are now often used in manufacturing of aerospace parts to reduce weight and maintain structural performance and reliability. Besides lightweighting, fiber-reinforced composites have another advantage over metals – failure of composites is not abrupt or sudden, but rather progressive. Even though progressive failure is advantageous from a life-performance point of view, it poses significant challenges when it comes to modelling and simulation. Another big issue related to usage of composite materials in various industries, including aerospace, is that their design has relied on empirical design practices, which are too costly. Therefore, efficient computational methods capable of reducing the overall cost associated with application of composite materials in industrial scale are highly desired. Accurate prediction of composite material performance requires a rigorous physics based approach to properly capture the microstructural behavior under various loading conditions. One of the main challenges is the modelling of multiple microstructural failure mechanisms, such as fiber break, matrix cracking and fiber-matrix debonding. In addition to that, modelling a full composite part with such great level of microstructural detail is not feasible, many times impossible, for standard numerical methods even when using modern high performance computing. In this work, a TRUE Multiscale Finite Element approach has been used to predict failure of carbon-fiber reinforced materials with the ability to consider all relevant microstructural failure mechanisms. Multiple 10-degree coupon tests, which is a non-standard test for continuous fiber reinforced composites, have been performed to validate the numerical approach. This test was selected with the goal to evaluate if the approach is capable of capturing de-bonding between fiber and matrix. It is shown that the computational models can efficiently and accurately capture microstructural details, including manufacturing-induced defects. Numerical results are compared to experimental data and total CPU time and memory are reported to show how efficient the TRUE Multiscale approach performed in this study.