Though the necessity and importance of quality guarantees for voice over IP networks (VoIP) are well understood and studied, not much has been done to evaluate and measure the quality of service (QoS) for VoIP in a practical commercial environment. This study focuses on an evaluation of media gateway performance, in terms of voice quality, affected by impairments of an IP network in a practical environment. To study critical elements that affect voice quality in general, two end-to-end VoIP networks were built in the Lucent Technologies, Next Generation Network (NGN) Interoperability Lab. Various IP network impairments, such as IP network delay, jitter, and packet loss were introduced into these systems for assessing the IP network impact on voice quality. The performance metric is the end-to-end voice quality. This paper presents the end-to-end VoIP test-bed architecture, the test configuration, the experiment methodology, testing tools, analytical results, and testing results. The performance results are viewed from a user's perspectives in terms of perceptual speech quality measure (PSQM) and speech latency. Finally, the paper points out the crucial factors that affect a successful VoIP network. Some possible remedies are suggested.