In this paper we explore a set of techniques to reduce the bandwidth in remote navigation systems. These systems, such as exploration of virtual 3D worlds or remote surgery, usually require higher bandwidth than the common Internet connection available at home. Our system consists in a client PC equipped with a graphics card, and a remote high-end server, which hosts the remote environment and serves information for several clients. Each time the client needs a frame, the new image is predicted by both the client and the server and the difference with the exact one is sent to the client. To reduce bandwidth we improve the prediction method by exploiting spatial coherence and wiping out correct pixels from the difference image. This way we achieve up to 9:1 reduction ratios without loss of quality. These methods can be applied to head-mounted displays or any remote navigation software.