Statistical identification of isochore structure, the variation in large-scale GC composition (proportion of DNA bases that are G or C as opposed to A or T), of mammalian genomes is a necessary requirement for understanding both the evolution of base composition and the many genomic features such as mutation and recombination rates, which covary with base composition. We have developed a Bayesian method for isochore analysis that we demonstrate to be more accurate than the commonly used binary segmentation approach implemented within the program IsoFinder. The method accounts for both fine-scale and large-scale structure. We adapt direct simulation methods to allow for iid samples from the posterior distribution of our model, and provide an accurate approximation to this that can analyze data from a chromosome in a matter of seconds. We apply our method to human chromosome 1. The resulting estimate of how GC content varies across this region is shown to be a better predictor of local recombination rates than IsoFinder, and we are able to detect regions consistent with the classic definition of isochores that cover 85% of the chromosome. We also show a measure of relative GC content to be particularly predictive of local recombination rates.