Analysis of diverse eukaryotes has revealed that recombination events cluster in discrete genomic locations known as hotspots. In humans, a zinc-finger protein, PRDM9, is believed to initiate recombination in >40% of hotspots by binding to a specific DNA sequence motif. However, the PRDM9 coding sequence is disrupted in the dog, raising questions regarding the nature and control of recombination in dogs. The sequence analysis of PRDM9 orthologs in a number of dog breeds and several carnivores shows that this gene was inactivated early in canid evolution. Patterns of linkage disequilibrium employing more than 170,000 SNP markers from almost 500 dogs to estimate the recombination rates in the dog genome show a good correspondence with an existing linkage-based map. Significant variation in recombination rates is observed on the fine scale, and over 4000 recombination hotspots are detectable. In contrast to human hotspots, 40% of canine hotspots are characterized by a distinct peak in GC content. A comparative genomic analysis indicates that these peaks are present also as weaker peaks in the panda, suggesting that the hotspots have been continually reinforced by accelerated and strongly GC biased nucleotide substitutions, consistent with the long-term action of biased gene conversion on the dog lineage. These results are consistent with the loss of PRDM9 in canids, resulting in a greater evolutionary stability of recombination hotspots. The genetic determinants of recombination hotspots in the dog genome may thus reflect a fundamental process of relevance to diverse animal species.