Within the field of cultural heritage restoration, experts are interested in the analysis of data describing the condition and history of ancient monuments. Data is usually distributed over many sites. VRML and Java technology, which are well-suited for describing geometrical models and data interaction over the Internet. Unfortunately, the poor quality of VRML real time rendering is a bottleneck for any analysis based on accurate image synthesis methods. Another problem in reproducing images with photorealistic rendering derives from the adaptation mechanisms of the human visual system. We describe a method and its implementation for providing high quality photorealistic image synthesis of ancient building materials. In this method, a network based Java application manages geometric 3D models of an ancient building to provide an editing interface and to manage high quality photorealistic snapshots. Simple 3D VRML data are enhanced with radiometric data derived by gathering measurements on the actual material taken from the site which is being reproduced. In the example presented in this paper, we have used actual measurements taken from the ancient Roman Aosta Theatre. A server-based optimised rendering application computes photorealistic images on radiometric data, that are subsequently applied as input to an algorithm simulating the human visual system perception. This latter phase is able to emulate most of the human adaptation mechanisms, including factors such as colour constancy, local lightness, and chromatic adaptation. The enhanced interaction with high quality images of the model through the Java application, allows a visual qualitative evaluation of restoration hypotheses. It also provides a tool that is able to show the final appearance of the model under assigned lighting conditions, as observed by a human being inside the virtual environment.