In this work, we extend the notion of generalized Hamming weight for classical linear block codes to linear network codes by introducing the network generalized Hamming weight (NGHW) for a given network with respect to a fixed linear network code. The basic properties of NGHW are studied. We further show that NGHW can be used as a tool to characterize the security performance of a linear network code on a wiretap network. We also introduce the notion of network maximum distance separable code (NMDS code) by extending the notion of Maximum Distance Separable code in classical algebraic coding theory. We prove that NMDS codes play an important role in minimizing the information that an eavesdropper can obtain from the network.