The use of Carbon nano tubes (CNT) can result in major improvements of manufactured thermally and electrically conducting plastic parts. Due to their high length-to-diameter (L/D) ratios, and their high conductivity, very low concentrations of CNT may be employed. This lowers the price, allows keeping the shape of a part and may even improve on the strength of a part, while significantly enhancing the conductivity. Enhanced conductivity requries a connected path of CNT through the part, i.e. that the network of CNT percolates [1]. The GeoDict / FiberGeo model of fiber networks [2] is combined with the GeoDict / Thermo- Dict [3] solver to predict the effective thermal conductivity and COGraph [4, 5] to predict the effective electric conductivity and percolation of CNT composites. The contrast of the thermal conductivity is up to 4 orders of magnitude and the contrast of the electric conductivity is up to 20 orders of magnitude. While ThermoDict works on 3d virtual tomographic images (vCT) of the composite, COGraph considers only the network of CNT and neglects the matrix material. ThermoDict has strenghts for smaller domains and lower contrast. COGraph works strictly above the percolation threshhold, can handle very large domains (as required by low CNT concentrations) and extremely high contrast (1e20) very efficiently and robustly. The predictions for different L/D ratios and varying degrees of anisotropy of the CNT fibers are compared with analytic models and experimental data for CNT-polymer compounds to ultimatelyl find optimal volume fractions and L/D ratios of CNT fibers for applicatrons.