The visualization and interactive navigation of 3D models, like virtual visits of museums, churches or even whole cities (cybercity models), are gaining a more and more increasing use and importance with time. This has been accomplished in last years thank to the augmented power of modern CPUs, graphic hardware devices and thank to a wider access to the web. Even in the survey field the interest towards 3D models has grown, given the advantages offered by ground-based laser scanners in terms of survey time and the pretty good quality of obtained 3D models. Indeed, several tests and applications have been performed so far all around world with such instruments: 3D data have been collected from objects of different shape and size, ranging from cultural heritage to land infrastructures, accuracy has been investigated and corresponding 3D models have been created using appropriated software. Beside issues related to the accuracy and the quality of resulting models, another interesting topic should be considered: the remote use of these 3D models, i.e. how they can be optimally transmitted over the web. To this aim we have developed a VRML split-browser, i.e. a 3D model visualization shared system based on two different server-side and client-side applications. The former executes the rendering of the scene and send it to the client, which is located on the remote PC of the user. In turn, the client provides the user with a graphical interface to explore the 3D model interactively. The system is based on image compression and projective transformations, which allow to share the 3D models over the web with limited throughput to the client. In this paper, we will present the frame prediction algorithm, based on a space displacement of the viewpoint and the way adopted to apply it in the remote visualization system.