When the same subject repeats a test under different conditions (e.g. evaluating two systems) there may be learning effects. One way around this is to have different subjects for each condition, but to have two or more subjects who are very similar (in gender, age, skills), and to allocate one from each pair (or matched group) to each condition. Matching subjects makes results more robust, requiring fewer assumptions, but typically has less statistical power than a within-subjects experiment.
Also known as matched users