As a new type of textile composites with broad application prospects, it is essential to study the prediction method of the mechanical properties of 2.5 D woven composites (2.5DWC). Currently, the most popular prediction method is to use a representative volume cell (RVC) for numerical simulation, so the reasonableness of RVC determines the prediction accuracy. However, many practical factors are ignored in the traditional periodic unit-cell model (UCM), such as the weft-layer-number (WLN), resulting in low prediction accuracy; while the full-cell model (FCM) in which the surface extrusion effect (SEE) and WLN are considered has the problems of complex modeling and high computational cost. To solve these problems, a triple-cell model (TCM) system is proposed, which includes four RVCs that are applicable to different WLNs, each of which is composed of different sub-cells (surface-cell, transition-cell, and inner-cell) which are categorized according to the characteristics of the actual weft yarn cross-section. Based on the progressive damage method, the stiffness, strength, and damage behavior of 2.5DWC with different WLNs are predicted, and the TCM prediction results are compared with the results of the experiment, the UCM, and the FCM. Compared with the experimental results, the prediction accuracy of the TCM is more than 8% higher than that of the UCM, and the difference between the prediction results of the TCM and FCM is less than 5%. Therefore, the proposed TCM system has the characteristics of high prediction accuracy, relatively simple modeling, and the applicability of any WLN.