A hybrid (signal-symbol) approach for detecting significant changes in imagery uses a signal-based change detection algorithm followed by a symbol-based change interpreter. The change detection algorithm is based on a linear prediction model which uses small patches from a reference image to locally model the corresponding areas in a newly acquired image, and vice versa. Areas that cannot be accurately modelled because some form of change (signal significant) has occurred are passed on to the change interpreter. The change interpreter contains a set of "physical cause frames" which attempt to determine if the change is physically nonsignificant (e.g., due to clouds, shadowing, parallax effects, or partial occlusion). Changes due to nonsignificant changes are eliminated from further consideration. If the physical cause of the change cannot be determined, it is passed on to an image analyst for manual inspection. Preliminary results of work in progress are presented. These results indicate that the methodology is extremely effective in screening out large portions of imagery that do not contain significant change as well as cueing areas which are potentially significant.