Raw files aren't really images as such, but the data straight off of the camera's sensor. I think of it as freezing the state of the camera immediately after the exposure has finished, and bundling it up into a file. The wikipedia article provides a fair amount of detail about the sort of thing that's recorded, but as a guide I'd say:
- All the levels readings from the sensor
- Some information about the sensor type (so the Raw converter knows how to interpret the sensor data)
- All of the settings currently applied to the camera
- The time and other EXIF-type meta-data associated with the image
- Possibly a JPEG thumbnail image
Bare in mind that there is no single Raw format, and all camera manufacturers are free to bundle up this sort of data in their own way.
I would imagine that because of how most sensors are built with a Bayer filter-type design, the data from the sensor just contains levels from each photo-site on the sensor. A bitmap, for example, would have combined the RGB sites to form a single pixel.