Some cyanobacteria have been shown to exchange genetic information under laboratory conditions, but it has not been clear whether such genetic exchange occurs in the natural environment. To address this, a population genetic study was carried out on the filamentous diazotrophic cyanobacterium Nodularia in the Baltic Sea. Nodularia filaments were collected from 20 widely distributed sampling stations in the Baltic Sea during June and July 1998. Allele-specific PCR (AS-PCR) was used to characterize over 2000 filaments at three loci: a non-coding spacer between adjacent copies of the main structural gas vesicle gene gvpA (gvpA-IGS), the phycocyanin intergenic spacer (PC-IGS) and the rDNA internal transcribed spacer (rDNA-ITS). The three loci were all found to be polymorphic in the 1998 population: two alternative alleles were distinguished at the gvpA-IGS and PC-IGS loci, and three at the rDNA-ITS locus. All 12 possible combinations of alleles were found in the filaments studied, but some were much more common than others. The index of association (I:(A)) for all possible pairwise combinations of isolates was found to differ significantly from zero, which implies that there is some linkage disequilibrium between loci. The I:(A) values for 16 out of 20 individual sampling stations also differed significantly from zero: this shows that the observed linkage disequilibrium is not due to pooling data from genetically distinct subpopulations. Monte-Carlo simulations with random subsets of the data confirmed that some combinations showed significantly more linkage disequilibrium than expected by chance alone. It is concluded that genetic exchange occurs in the natural Nodularia population, but the frequency is not high enough for the loci to be in linkage equilibrium. The distribution of the 12 genotypes across the Baltic Sea was found to be non-random, but did not correlate with temperature, salinity or major nutrient concentrations. A significant relationship was found between the gene diversity among filaments at each station and the distance of the station from the centre of the sampling area: possible reasons for this trend are discussed.