Paul thinks this could be due to the following jdk bug: http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=4648098 We don't put the vfs into the system class loader when running the server or under eclipse so I don't see that this directly applies. Since the bug is old, it could be that the handlers must be int he application class loader and mvn test harness is creating a child class loader of that, and the handlers are only visible from the thread context class loader, and so some form of this bug applies. We need to see what class loaders exist when the first call to resolve a vfs* url is made. This will require hacking the jdk source and running with -Xboostrap/p:/path-to-jdk-source-classes to override the handler loading code. View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4032409#... Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&a...