The above diagram shows the relationship between photosite pitch/area and resolution, with a few recent cameras located. The "SS" is the proverbial "sweet-spot" that people talk about. This diagram is based on facts, but I don't have proof that the obvious conclusion (ie, that IQ falls off as photosite pitch gets really large and really small) is true. I may just not have the time or I may not know how, but just because I don't/can't prove it doesn't make it wrong. It seems obvious that a camera with big photosites must be big physically. It also seems obvious that extremely tiny photosites don't have very good dynamic range and have rather high noise. It doesn't seem like a big leap to conclude that somewhere in the middle things are better.

Note that the NUMBER of pixels isn't a factor in spatial resolution. A camera with a 6 um photosite pitch can have any (reasonable) number of pixels and it will still record the same resolution. Whether the resultant image looks good depends on a lot of other things.

An analogy: Back in the film era, Kodachrome 16 ASA film had a predictable (high) quality, regardless of the size of film used. A tiny Minox "spy" camera could use Kodachrome 16 ASA film. A huge Kodak Medalist 2 1/4" x 3 1/4" camera could also use Kodachrome 16 ASA film. The "quality" of the pictures produced by both these cameras were comparable, as long as they used that same Kodachrome 16 ASA film. The big camera produced a bigger image and the little camera produced a smaller image with the same IQ...as long as we compare equal areas of film.

If we load both cameras with Tri-X and push development to get 1600 ASA sensitivity, both cameras produced lower quality images...ones with much more grain and less resolution. These IQ issues were more noticeable with the small format...the big format camera produced images that could be enlarged to a greater extent, but that didn't mean that the film was somehow better when used in the big camera. The sweet spot was, of course, in the middle...a 35mm camera using ASA 100 film.

Unlike film cameras (which use photons to trigger a chemical reaction) digital cameras count photons and in the physics of photon counting, the noise in the signal is equal to the square root of the number of photons. The number of photons are affected by the aperture, the exposure time, the illumination level, the packing fraction of the sensor (what % of the total available area is occupied by the photodiode), the total available area, and the capacity of the electron well associated with each photosite. Many of these variables can be ignored when pondering the effect of photosite size. Roger Clark has found that the signal-to-noise ratio is proportional to the photosite pitch for actual cameras of a given "generation". We can use that to simplify a very complex subject.

I tried a simple experiment to visualize the differences in sensors as we change the number of pixels. In order to eliminate most of the variables, I "created" an imaginary camera...a camera with a 1/2.5" sensor (that's one 5.76mm wide). I gave it 9 different numbers of pixels. I created vector artwork and added noise according to the pixel-pitch. I also blurred the result to simulate what camera manufacturers do to minimize noise. These may not be exact for every camera, but my imaginary camera has very typical performance. This table shows the data for the 9 pictures:

File	Photosite Dims				5.76			Noise	Blur			Dynamic
#	X	Y	DPI	MP	Pitch, u	Area	SNR	%	%	Black	White	Range
1	2940	1960	1400	5.76	2.0	4	2.0	15.0	2.6	16	236	14.8
2	2520	1640	1200	4.13	2.3	5	2.3	12.9	2.2	13	240	18.5
3	2100	1400	1000	2.94	2.7	8	2.7	10.7	1.8	12	244	20.3
4	1680	1120	800	1.88	3.4	12	3.4	8.6	1.5	9	246	27.3
5	1260	840	600	1.06	4.6	21	4.6	6.4	1.1	7	249	35.6
6	840	560	400	0.47	6.9	47	6.9	4.3	0.7	5	251	50.2
7	420	280	200	0.12	13.7	188	13.7	2.1	0.4	3	253	84.3
8	210	140	100	0.03	27.4	752	27.4	1.1	0.2	2	253	126.5
9	105	70	50	0.007	54.9	3009	54.9	0.5	0.1	1	255	255.0

This composite shows the 9 pictures:

The numbers below each image reference the above table.

You can see differences, but I'll bet it's difficult for you to decide which one looks best! One corollary to this riddle is that for a camera with a big sensor, the differences are subtle...a big sensor is going to give you good pictures regardless of the number of pixels it has (at least no manufacturer has yet given us a really terrible, large-sensor camera). That's why I picked a small sensor...when I did the experiment with a big sensor, the results looked almost the same on a cumputer monitor. Were I to do this with large, printed images (instead of your monitor), it would show similar differences even with a large sensor. Note that it's hard to choose between high noise and low resolution...except at the top where the resolution is terrible and the bottom where the dynamic range is very low! Out in the middle, they all look pretty good. While this doesn't prove that there is a sweet spot, it at least demonstrates that there is a sweet area.