In modem computational mechanics, typically, but not exclusively using Finite Element techniques, solution parameters are tabulated for discrete points in the problem domain. A typical large dynamic analysis may have a complex 3D geometry modelled with a million elements, with results tabulated over a large number of time steps. From this enormous 4D data set, the stress engineer must select appropriate cross-sections in order to visualise the most significant features. The choice of such cross-sections is usually subjective, based on experience and engineering intuition. Typically, the data set will be manipulated in a high specification server, and the graphical information will be pushed down the network to engineer's local machine. However, there are three practical disadvantages: the graphical computation can be time consuming - each cross-section through the model can take several minutes to load, selection of views is by trial and error, and interactive demands on the server interfere with its performance on the other large finite element analyses, which it is processing. For the purpose of exploring the solution domain, faster computer response is desirable. A lower fidelity model of the post-processing data may be acceptable for this purpose, and the graphical display of a model of sufficiently reduced size could be managed by the local workstation. Wavelets have been used with great success for 2D image compression, with compressions to 5% of original data still giving visually acceptable images. This leads us to expect that for 4D compression, a reduced model of 0.25% of the original size may be practical. Wavelets have the advantage over other methods of image compression in that sharp contrasts are preserved. It is, therefore expected that areas which would be of interest to the stress engineer, for example rapid changes in the stress field, high deformation rates, etc., would be preserved in high fidelity. This feature of wavelets may be further harnessed, to give a basis on which some of the most significant cross-sectional views may be automatically computed and presented to the engineer at the beginning of the post-processing stage. This paper will explore the amount of compression practically achievable, and describe the algorithms required to achieve this. The results will be shown in the form of a demonstration analysis, with both high and low fidelity results presented side by side.