This review paper discusses known methods and their prediction accuracy for the thermal conductivity calculation as well as their application to molecular liquids, polymers, and carbon nanotubes. Particular attention has been paid to the influence of simulation parameters, size-effects, and force field on the thermal conductivity calculated. The simulation parameters, such as the use of thermostat, exchange period in the algorithm, and atomic vs. molecular exchange, have the lowest impact on the calculated value. Variation of simulation parameters in reasonable ranges results in the mutual deviations of thermal conductivities within the respective error bars only. Size effects can be avoided, but should be considered for each particular system. For molecular liquids as well as for polymers, a simulation box of several nanometers length seems to be reasonable, whereas, for carbon nanotubes even several hundred nanometers are not enough due to the divergence of the thermal conductivity with the tube length. The choice of the force field appears to be most decisive for the calculated thermal conductivity. Correctly tuned united-atom models together with bond constraints lead to significant improvements in the prediction accuracy.