Estimates of allelic recombination frequencies are shown to have coefficients of variation of between 20 and 40%. In Coprinus this is true of both high and low recombination frequencies and is also true when the alleles involved show marker effect. This variability is not confined to Coprinus but is a general feature of both meiotic and mitotic allelic recombination. Experimental errors do not make a major contribution to the observed variation althought it has the nature of a sampling variation. It is suggested that the variation arises from the diversity of ways in which the initial errors introduced by hybrid DNA formation can be resolved during the excision-repair stages of recombination. If the enzymes responsible for these processes are present in low concentrations then much latitude can be anticipated in the way the same errors are dealt with by separate, though isogenic, diploid or dikaryotic organisms. Each separate cross is thus interpreted as providing an estimate of the recombination frequency which is but a sample from a varied population of possible estimates of the same recombination frequency. Each pair of alleles exhibits a recombination frequency which, within the statistical boundaries of the general variation, is sufficiently reproducible to be described as a characteristic of them. Combinations of allelic recombination frequencies derived from pair-wise crosses fall into patterns that are sufficiently consistent for allele maps to be drawn; and, providing a sufficient number of replicate crosses have been analysed, the allele map can be shown to be statistically soundly based. Two marker effect situations are examined. One causes reduction of recombination frequency and is probably intrinsic to the mutant site itself, the other causes enhancement of recombination frequency and is due to a factor or factors distinct from the allelic mutant site in the strain in which it was first identified. When intercrossed the two effects counteract one another.