Random network for Lanchester combat, a novel theoretical model for combat is put forward based on the definition of random mapping. The model is an ER random sub-graph of directed bipartite graph, or a series of random mapping with the decreased size of mapping domains, in which the relationship of local interaction and the system behavior have been formally and explicitly represented. The results obtained mainly by simulation and analysis show that the efficiency of the whole combat network follows the Lanchester's square-law, and that the topological structure of the network generated by the model is heterogenous, following exponential-law of degree distribution. Finally, the influence from the factors of globe information on the efficiency of the whole warfare system is studied through data experiments for both “symmetric” and “asymmetric” combat cases. It ends up with a discussion on the significance of network modeling in the combat modeling theory, experiments and methods of calculation.