ISO: International Standards Organization
ASA: American Standards Association

ISO and ASA are essentially the same thing - it is an agreed upon value for the sensitivity of film or imaging surface.  This allows us, for a given scene, to use the consistent shutter/aperture combinations regardless of the imager type.

Electronic devices all have an optimal level where they function the most efficiently.  Amplification above this baseline value decreases the signal-to-noise ratio.  In digital photography this noise shows up as "grain."  The more a weak signal is amplified, the more difficult it becomes to distinguish actual information from background noise - the image becomes "grainy."

The image to the right is a 100% crop.  Pixels are depicted in a 1:1 ratio - each sensor pixel is represented by a single pixel in the image (with the appropriate Bayer interpolation).  This view allows us to analyze noise patterns with excruciating exactness.

If the image were resized, noise characteristics change due to the averaging and interpolation that occurs with resizing.  Resizing down decreases visible noise by averaging noise out with neighboring pixels.  Upscaling an image will distribute the noise pattern as the pixels themselves become bigger and interpolated (combined) with neighboring pixel information.

This crop contains 1/10th of the actual image map.  It is an ideal size as it is easily transmitted over the internet and allows for side-by-side comparisons.  We can use this technique to visually see the difference between sensor performance characteristics.

The above image was taken at ISO 200.  Move the mouse cursor to see the same image taken at ISO 1600.  Click on the image to see a side-by-side comparison.

Noise reduction techniques are available, but tend to confuse high frequency data (read: detail) for background noise.  The net result is noise reduction robs detail from images.

As stated earlier, CMOS sensors are inherently noisier than CCD sensors.  Manufacturers combat this with different approaches.  Canon, for example, applies noise reduction algorithms to the data coming off the sensor before it is actually recorded.  Kodak, simply records all of the data, including noise, and deals with it using software.