Visual Preference Surveys are one great use for a collection of images. A Visual Preference Survey takes the pulse of a group of citizens concerning their preferences between various pairs of objects. We use them to test public preference for various languages (styles) of architecture, but they could also be used to test for many other object types as well.
To begin, select two sets of images... some use 50 pairs, others use 100, although the number isn't critical. Rather, there should be enough pairs to show trends, but not so many that they cause participant fatigue. You'll project each pair of images side-by-side, and participants will choose which one they prefer.
Each pair of images should show the same building type, but in two different languages of architecture. In other words, comparing a house to a gas station will almost always favor the house, so that result isn't useful. Also, a cottage often gets better ratings than a large house, so the size of the buildings should be similar. Also, you should control for image externalities. In other words, if one image has a blue sky, the other should as well, as a blue-sky image almost always rates higher than one taken on a grey and cloudy day. The bottom line is that you want to make the architecture the primary difference between the images, so that the results are meaningful.
Participants are given a simple scorecard, and are instructed to check the right box if they prefer the right image, or the left box if they prefer the left image. Each image pair is numbered, as are the checkboxes on the scorecards. When the survey is complete, the results are tallied in a database where they can be analyzed.
The results are often striking. Whereas we as architects can debate endlessly over architecture, the non-architects are usually quite decisive. It's really a technique we should use more often, IMO.