This study is devoted to assess sex ratio variation among 33 populations of the gynodioecious Beta vulgaris ssp. maritima in Brittany (France) and to explore the causes of this variation. We showed that three different CMS (cytoplasmic male sterility) cytotypes occurred in populations, but strongly differed for their frequencies and the frequency of their associated nuclear restorer alleles (which counteract the effect of CMS and restore male fertility). No correlation was found between CMS and restorer frequencies within populations, which has been previously interpreted as a result of stochasticity. However, neutral genetic variation did not indicate recent population bottlenecks in studied populations. Moreover, no significant correlation was found between female frequency or variance and current population size. Consequently, stochastic processes could not be the major cause of sex ratio variation. Alternatively, empirical estimations of the variation of females, CMS genes and nuclear restorer allele's frequencies were compared to theoretical predictions based on a frequency-dependent selection model of gynodioecy. In particular, we showed that an absence of correlation between CMS and restorer frequencies could also occur without stochasticity. The large variation of sex ratio in Beta vulgaris could thus be explained by frequency-dependent selection acting on CMS genes and restorer alleles.