Packet duplication is discussed as a means of incre asing network re liability in an environme nt whe re packe t loss exists. Sever al methods of routing the duplicates are prese nted, one of which-the stnumbering--is shown to have the combined adva ntage of using disjoint paths and more eve n utilization of network resour ces. An additional mecha nism, delibera te packe t elimination, is introduce d as a means of controlling congestion that may re sult, in par t, fr om the duplication. A compre hensive model is defined enc ompassing the proce ss of pac ket duplication together with both forms of pac ket elimination. Within this model a cost func tion, based on ave rage pac ket delay, is defined. A quasi-static distributed algorithm is develope d that is optimal, dea dlock-fr ee and loop-fr ee. Extension of the model to include pac ket retr ansmission is consider ed.