The method of PBarwick is the way to go if you only have to handle one instance of an object, however it can get very tricky if you have to handle a lot of instances of the same object, as you will have to use for each loops to handle the correct assignment of the collision objects to their parents.
I strongly recommend this tutorial about setting up so called "embedded" collision detectors.
http://www.diybandits.com.au/MMF/tut...embedded1.html
What it basically means is that you have your active object --> switch to an embedded detector animation (the shape you want) and check for collision --> switch back to your displayed animation before the renderer triggers (end of your frame loop). If you really learn how to use it will allow advanced AI programming and a lot of other fancy stuff!
Unfortunately there is one big problem: The method is broken since build 285.0 (beta). It seems that CT changed the way animations are restored causing animations to freeze if you switch to an embedded detector and back within one frame loop. I posted it in the bugbox, but I don´t expect to get this fixed soon.. However you are fine if you use the last official release build 284.10
Hope that helps!
http://bugbox.clickteam.com/issues/3421