First step is always to try to reproduce it on our PCs. If that's not possible, _BugReport.txt can sometimes reveal the problematic piece of code. Another option is to provide the reporter with a special assert build, which might show some errors and problems in the application logic early on. If that's unsuccessful too, we need to find at least a correlation with something that the reporter has in their system and we don't. If we cannot find that, as a last resort we can connect remotely to the client's PC, with their permission, and we can debug the issue there.