- Set both cameras to full manual mode, RAW (no processing-- the jpg algorithms will interfere)
- Place one camera on a tripod, aimed at something relatively innocuous (a wall of books). Do not aim the camera at something that moves (person, pet, etc)
- Shoot using some group of settings (f/8.0, 1/60 shutter, 100 iso, 35mm, or whatever produces a reasonable exposure-- if you're not used to doing this, you may need to experiment to get something good on the back view of the camera). Write down these settings, or remember them.-+-
- Shoot the same scene with the same camera on auto mode.
- Repeat steps 2-4 with the other camera.
You now have two sets of images. One set is where you have controlled the settings of the camera, and these images should show the difference of the sensors directly. The other shows the differences in the auto settings.
If your rebel is not producing a 'better' (more noise-free, better color rendition, less blurring between pixels) image when using manual settings, then either the s90 is just that much better (doubtful) or your camera is broken. You can perform the same test with another rebel, if you have access to one, just to see if the cameras are at all different (they should not be).
Now, it may be possible that on auto settings, the s90 is producing a 'better' image than your rebel. 'Better' is a very subjective term, and what looks good to your eye may not look good to others-- for instance, my wife loves brighter images, and I tend to the darker. If you find that on auto settings the s90 produces what you want out of an image, and your rebel does not, you can either learn how to make the rebel do what you want or you can just use the s90.
-+- remember that both cameras have different meanings for their zoom settings. That is, 35mm on your rebel will probably not be the same image as on your s90, because they have different chip sizes in them. What you want is the same angle of view. That's why I suggest a bookshelf-- you should have a good idea of where to zoom to with both cameras so that the images are roughly similar.