If you take a sample of fragmented genomic DNA, which contains < 1 genome in total, not all possible genomic fragments will be represented, some sequences will be missing. If you then take several such samples (the panel), any single marker will only be present in a subset of the aliquots.

Using PCR to detect the presence or absence of any single marker within the different aliquots ( typing a marker), we can get a picture of the distribution of this sequence within our samples (as in the photo on the right).
If two markers are close together within the genome, they will have a higher chance of being in the same genomic fragment. Therefore, they will have a higher likelihood of co-segregating with each other into the different aliquots.

So, if we test the same set of samples as above with a second marker, which lies close to the first, we expect to see two patterns of distribution which are extremely similar (comparing the photos above and left).
A third marker, which lies a bit farther away on the genome, may occasionally fall within the same DNA sampling as the other two, but will frequently segregate into completely different aliquots. It's distribution pattern, within the same panel, will show similarities to the first two but also some differences (compare the photo on the right to the others)

By statistically analysing the similarities and differences between the patterns of many markers, on the same panel, we can infer how close or how far away from each other they are within the genome, and hence their order, to produce a map.