This is a hot topic in the amatuer community and has lead to many heated discussions on the internet and compuserve astroforum. I have had plenty of experience with CCD camera's over the last 5 years which include use of the ST6 with large 23 x 27 micron pixels and the ST7 with 9 micron pixels. Below, I discuss a practical view of which pixel size is best and please be aware this is my own opinion!
Scientific facts:
The resolution of a CCD image is determined by how "tight" star images are formed on the CCD chip and is determined by many factors which include accurate focus, accurate guiding, optical quality and local seeing conditions.
Light is converged onto the CCD detector and the size of this area is called the spot size which equates to the airy disk. The formula for the radius of the spot size is: r = 1.22xwavelengthxf/D, where f = focal length and D = aperture diameter of the telescope. Hence the spot size decreases linnearly for increasing diameter size of the primary mirrior or lens. This makes sense as there is no way a 4 inch telescope can equal a 200 inch telescope in terms of resolution. However, this only holds true for apertures greature than 4 inches, as for small pertures the spot size is dependent on the focal ratio.
The nyquest sampling theory states that at least 2 pixels must cover a star image in order not to lose information. Imagine two stars covering adjacent pixels and each star only covers one pixel. This results in a rectangular shaped object (combining both stars) with no discernable gap between them. If, however, both stars are over-sampled, then light from each star will be concentrated in the center with less to adjacent pixels (point spread function). Although there is still no gap between stars, two stars will be seen shaped like a pair of binoculars.
Therefore pixel size is dependent on focal length (assuming your telescope aperture is over 4 inches in diameter) and at least 2 pixels are needed for a star image. Remember not to over-sample otherwise stars will appear as a "blob" rather than a nice round point source. My seeing conditions of 3-4 arc seconds (typical for most amatuers) mean I need the pixel size to correspond to 1 1/2 - 2 arc seconds. Unfortunately CCD camera's with smaller pixels do have their drawbacks. Firstly they are less sensitive and longer exposures are required and secondly the full well capacity is reduced.
Pixel scale in arc secs = (205 x p)/fl where p = pixel size and fl = focal length in mm. For example, using the ST7 with 9u pixles, my 5 inch AP scope at F6 produces 2.2 arc secs per pixel and my 12 inch LX200 at F10. 0.6 arc secs per pixel (over-sampled). With the ST6 and 27u pixels, the 5 inch AP scope at F6 produces 5.5 arc secs per pixel (well undersampled) and 2.2 arc secs per pixel on the 12 inch Meade at F10, confirming my own practical experience that the ST6 is better suited to the 12 inch Meade and ST7 to the 5 inch refractor (see below).
My experience
Accurate guiding and focus are discussed on these Webb pages and with the equipment I have, do not find them to be a limiting factor in resolution. Optical quality is important and both of my telescopes (particularly the astrophysics refractor) perform well on benchmark testing and only with exceptional seeing conditions, could better optics be of value.
By far the most important factor is the local seeing conditions and this is usually about 4 arc seconds where I live. Rarely, I have experienced 3 arc seconds and the resolution in CCD images obtained is then MUCH improved In practical terms, I find the ST7 with 9u pixels is superbly matched to the 5 inch Astrophsics refractor and the 12 inch LX200 at F3.3 (with Optec focal reducer) for my typical seeing conditions. The ST6 on the other hand is better matched to the longer focal length of the 12 inch Meade LX200 at F10. IF, I
had superlative seeing conditions of 1 arc second, then the ST7 would be well matched to the 120 inch focal length of the 12 inch LX200. You can see these results from images on these Web pages.
Below is an image of M33 taken with the 5 inch astrophysics refractor at F6 and an ST7 camera in 9u pixel mode. The image on the left is unprocessed and on the right, I reduced the resolution by 9/25 to simulate 25u pixles and then increased the image size to match the original. There is clear loss of resolution, in particular look at the stars in the outer spiral arms and close doubles which are lost. The stars are also blocky and unsightly. Which do you prefer??
If there is doubt about which camera to buy, I would always recommend one with small pixels as this will yield high resolution images when seeing conditions allow and on chip binning can effectively increase the size of pixels for those poor nights. CCD camera's with larger pixels than necessary are probably more satisfying to use for the beginner as they are more forgiving on guiding and because of increased sensitivty, shorter exposures produce images with a good signal to noise ratio. However, with experience you will long for better resolution.