The conservation of a crop's wild relatives as genetic resources requires an understanding of the way genetic diversity is maintained in their populations, notably the effect of crop-to-wild gene flow. In this study, the amount of differentiation between natural and cultivated populations of Medicago sativa was analysed using random amplified polymorphic DNA (RAPD) markers and an extension of the AMOVA procedure adapted to autotetraploid organisms. Simulations of structured populations were performed to test whether AMOVA provides estimates of population structure in autotetraploids that can be directly compared to those obtained for allozyme data. Simulations showed that straight phi-statistics allow a good estimation of population differentiation when unbiased allelic frequencies are used to correct the conditional expectations of squared genetic distances. But such unbiased estimates can not be practically guaranteed, and population structure is notably overestimated when some populations are fixed for the presence of amplified fragments. However, removing fixed loci from the data set improves the statistical power of the test for population structure. The genetic variation of 15 natural and six cultivated populations of M. sativa was analysed at 25 RAPD loci and compared to estimates computed with allozymes on the same set of populations. Although RAPD markers revealed less within-population genetic diversity than allozymes, the quantitative and qualitative patterns of population structure were in full agreement with allozymes. This confirmed the conclusions drawn from the allozymic survey: crop-to-wild gene flow occurred in many locations, but some other mechanisms opposed cultivated traits to be maintained into natural populations.