Crop genebank managers must be concerned with preserving low-frequency alleles that are vulnerable to being lost from the genebank because they are not sampled as seed increase parents. Traits conferred by such alleles are also vulnerable to being overlooked because they are not sampled in plants evaluated for breeding traits. Mathematical models and recommendations that follow from hypothetical vulnerable alleles within single populations have been made by various researchers. This work was undertaken to empirically detect the prevalence of such alleles in the US Potato Genebank by extrapolation from RAPD markers. Two outcrossing species known to be very heterogeneous were selected, Solanum jamesii Torr. (2n=2x=24) and Solanum sucrense Hawkes (2n=4x=48). RAPD band frequencies at a total of 83 marker loci within each of 15 populations were assessed using 24 plants per population. For both species, about 10% of within-population loci had vulnerable alleles (banded plant frequency <40%), considering just the particular population in question. However, about half of these were fixed in at least one other population (making that marker invulnerable to loss). The highest within-population banded plant frequency for loci not fixed in any population was, with one exception, always >40%. So, while several alleles may be vulnerable within a given population, these were nearly always fixed or nearly fixed (invulnerable) in another population. This suggests that the safest strategy for maximizing conservation of alleles is to apportion available resources over many separate populations, rather than adopt a preservation protocol so rigorous that it can only be practically applied to a limited number of populations.