The effective thermal conductivity enhancement of carbon fiber composites is investigated in this contribution using a three-dimensional numerical method. First a more realistic three-dimensional distribution of fibers dispersed in a matrix phase is reproduced by a developed random generation-growth method to eliminate the overrated inter-fiber contacts by the two-dimensional simulations. The energy transport governing equations are then solved through the three-dimensional structures using a high-efficiency lattice Boltzmann scheme. The resultant predictions agree well with the available experimental data. Compared with the existing theoretical models, the present method does not depend upon empirical parameters which have to be determined case by case, so that it is useful for design and optimization for new materials, beyond prediction and analysis just for existing composites.