Research on mutualism seeks to explain how co-operation can be maintained when uncooperative mutants co-occur with cooperative kin. Gains and losses of the gene modules required for co-operation punctuate symbiont phylogenies and drive lifestyle transitions between cooperative symbionts and uncooperative free-living lineages over evolutionary time. Yet whether uncooperative symbionts commonly evolve from within cooperative symbiont populations or from within distantly related lineages with antagonistic or free-living lifestyles (i.e., third-party mutualism exploiters or parasites), remains controversial. We use genomic data to show that genotypes that differ in the presence or absence of large islands of symbiosis genes are common within a single wild recombining population of Mesorhizobium symbionts isolated from host tissues and are an important source of standing heritable variation in co-operation in this population. In a focal population of Mesorhizobium, uncooperative variants that lack a symbiosis island segregate at 16% frequency in nodules, and genome size and symbiosis gene number are positively correlated with co-operation. This finding contrasts with the genomic architecture of variation in co-operation in other symbiont populations isolated from host tissues in which the islands of genes underlying co-operation are ubiquitous and variation in co-operation is primarily driven by allelic substitution and individual gene gain and loss events. Our study demonstrates that uncooperative mutants within mutualist populations can comprise a significant component of genetic variation in nature, providing biological rationale for models and experiments that seek to explain the maintenance of mutualism in the face of non-cooperators.