Bobolinks have dialects that can be defined in terms of syllable structure and syllable order of the two songs typical of their repertories. Songs for each dialect are distributed over large but discrete geographical areas that correspond with disjunct patches of nesting habitat. Four of the nine dialect areas described in this study show evidence of subdivisions characterized by consistent differences in the dialect songs. Males singing non-local songs represented about 25% of the population in all dialect areas. Non-local singers almost never shared their songs with each other. In some cases, songs of non-local singers were the same as the songs of nearby dialects, suggesting that these males had dispersed between dialects on the study site. If non-local singers are assumed to be immigrants from other dialects, then most of the non-local singers encountered have dispersed from dialects off the study site, usually a distance of more than 20 km. Dialects were stable in geographic area and in song structure over the four years of the study. No individuals were observed changing song types either within or between years, although no individually marked males singing non-local songs were observed in more than one year. A few males siging local dialect songs made minor changes in syllable shape between years. A local dialect male first recorded as a known first-year male (born the previous summer) made major changes in syllable shape and order in his songs between his first and second breeding seasons. Overall, dialect songs were stable within and between years. The dialects on the study area were probably founded 80-100 years ago, following deforestation of the study site and immigration from other bobolink populations. Syllables were shared among a number of the nine dialects on the study site. These similarities may reflect a sharing of songs by the founders of the dialect populations followed by divergence among dialects due to independent cultural evolution within each dialect. The observation of subdivision in several of the study site dialects supports this pattern of dialect relationships. Alternatively, similarities among dialects may be explained by independent founding of each dialect followed by exchange of song material between dialects by immigrants. Inferred dispersal between dialects by males singing non-local songs supports this pattern of dialect relationships. The two models of relationships are not exclusive. Similarities between dialects could reflect both shared cultural ancestry and transmission of syllables between dialects. Song sharing in bobolinks was correlated with greater nest success. An alternative hypothesis, that reproductive success was associated with size, was not supported. To investigate how differences in breeding success might arise, differences in arrival date, territory quality, and residence period were examined in relation to song sharing. These three variables were all strongly correlated with breeding success. Song sharing males arrived earlier and were resident for more days than non-sharing males. Differences in territory quality related to song sharing appeared to be due to the earlier arrival of song sharing males. Two models proposing individual fitness benefits of song sharing, the social adaptation model and the genetic isolation model, were supported by the association between song sharing and reproductive success. Several other predictions of these models were tested using observations on dispersal patterns, song learning, dialect size, relationship of dialects to habitat patches and stability of dialects. Neither model could be rejected using the available data, but several important predictions of the genetic isolcation model were not tested in this study. A new hypothesis, the social experience hypothesis, predicts that response to playback of unshared song is dependent on prior individual experience with unshared song. This prediction was supported by the results of a playback experiment on bobolinks. The social adaptation and genetic isolation song dialect model predict that males should respond more strongly to unshared song than to shared song. This prediction was also supported by the results of the song playback experiment, although only when test males had experience with unshared songs. The importance of experience in recognition of unshared songs may be a result of poor signal generalization ability. Males captured between ages 33-65 days developed songs very similar to those of the dialect areas where they were captured, even in the absence of access to adult song models at any time after capture. One captive male had apparently dispersed from a nearby dialect before his capture at 65 days of age. This male did not develop the songs of the dialect where he was captured, but instead developed songs like those of a dialect 7.5 km from his capture location. Males captured in different dialects and housed together did not copy song material from one another during song development. There was evidence that males from the same dialect did copy each other to a slight extent. These results do not provide strong evidence for song learning in bobolinks but are concomitant with this process.