Nikon D800 official release, 36MP..

[ChadS]

@Rocco You've got to decide: same framing or 100% crop?

Let's say you have the identically-framed portrait of a model on a 10 MP and 18 MP 1.6x sensor. They'll look mostly the same if you put it up on the computer screen.

Let's say that if you zoom in on the model's eye you could count the number of pixels that make up that part of the image. For the 10 MP sensor that might be 10,000 pixels whereas that would be 18,000 pixels. So at this level (all other things being equal) the 18 MP image would look better. However, if you zoomed the 10 MP image so that you saw 2 MP worth of pixels on the screen and zoomed the 18 MP image so that you saw 2 MP of pixels on the screen you'd actually see less of the image in the second case (it would appear to be more magnified).

Now, at this higher resolution you are approaching (or well past) the ability of a lens to resolve detail with sharpness. So you will think that the second image looks softer than the first only because you're zooming in on the softness of the lens when in fact the second exposure can have more detail (can - not will). This is all for perfect sensors. Throw in less light per pixel for the 18 MP and you get more noise - hopefully balanced by better generation amplifiers and A/Ds. Throw in the vagaries of AA filters and you've got a convoluted mess.

The only way to know if you've got a softer sensor or not is to use something like the USAF resolution chart and shoot both. If you do that with the 40D and 7D you'll almost certainly be able to resolve tigher linepairs with the 7D.

Edit: noise and pixel size and statistics...

Ideally we'll have so many photons hitting our sensors that we don't have to amplify the signal very much and we get an overwhelming signal. Grand - but not practical. As we have to amplify the signal more (read: higher ISO) more noise is introduced into the signal as a consequence of the reality of physics of amplifiers. You've also go to consider the the number of photons that reach a sensor for a particular intensity is a statistical process. Even shooting a uniform white field you're not going to get exactly 10,000 photons of the same energy level on each pixel. Some will get more some with get less. However, the more photons there are the less this variation represents as a fraction.

As our intensity drops the number of photons registered by each pixel is a lower number. The pixel-to-pixel variation is therefore a larger fraction of the overall. Now, since the number of photons that strike a pixel is proportional to the area of the pixel the pixels on a 10 MP sensor will be larger than that of an 18 MP sensor. This is also why FF cameras of the same # of pixels have less noise than APS-C sensors. I'm simplifying because there are some other things in play like quantum efficiency of the sensor, coverage fraction, etc. but the main idea holds.

What John was saying is that if you Photoshop to reduce the 18 MP image to a 10 MP image using some sort of resampling the reduced-pixel image will appear less noisy than the original - simply because the pixel-to-pixel variations will be averaged out to a certain extent.