Prediction of void formation in the resin transfer molding (RTM) process will provide useful information such as the inhomogeneous microstructure and the mechanical properties of molded composite materials. This study aims at developing a numerical procedure to predict the resin flow with considering the microstructure of fiber preforms. Flow of liquids in imitated fiber bundles was first observed by using a transparent mold. The structure of the imitated fiber bundle was modeled by the moving particle semi-implicit (MPS) method, and liquid flow to the bundle was analyzed taking into account the surface tension and wettability as a potential force. Impregnation behavior depended on liquid properties, and the predicted process of void formation agreed well with the observation. This result indicated that the void content varied by the molding conditions such as the injection speed and liquid properties. Those molding conditions can be integrated into the capillary number (i.e., the ratio of the viscous force to the surface tension), and influence of this parameter on the void content was investigated analytically. The microscale flow analysis provided an appropriate molding condition to reduce voids with considering the fiber mat material and the resin.