I've made this statement repeatedly in the forums and thought it was about time to offer some solid proof.

I shot 10 photographs of my standard "white" target at all ISOs and without changing any other settings.
I analyzed the raw data with a program that computes (among other things) signal strength, noise, and signal to noise ratio.

As expected, the signal strength is "perfectly" linear with respect to ISO.
This is because the amplifier between the CCD and the Analog to Digital converter is very clean and does not introduce any measurable noise.

As expected, the noise is also "perfectly" linear with respect to ISO.

Now you may be saying, "see, noise is higher at higher ISOs!"; but the noise is the same, it is only amplified at higher ISOs and appears larger.

The number that really matters is the signal to noise ratio.
Since the signal and noise are both "perfectly" linear with respect to ISO we can expect that the signal to noise ratio will be constant.

The slight variation in these values is statistically insignificant.
The dashed line is the average value and the dotted lines are 2 standard deviations above and below the average.

What does this mean?
It means we should not worry so much about increasing ISO when necessary.
And when we do increase the ISO we shouldn't "blame" the higher ISO for the noise.
Also, we know that long exposures have increased thermal noise; we should probably increase ISO instead of (or perhaps in conjuction with) increasing exposure times to over 1 second.

Bill

Visit me, info and galleries at: Nikon Photos and Information

recently i discovered this the hard way.
while at kearney, NE last week i forgot to decrease the ISO from 1600 once the sun had risen. below is a cropped (not resized) image. i had expected more grain at that speed!

Sarge.