The environmental dependence of galaxy clustering encodes information about the physical processes governing the growth of cosmic structure. We analyse the baryon acoustic peak as a function of environment in the galaxy correlation function of the Baryon Oscillation Spectroscopic Survey CMASS sample. Dividing the sample into three sub-sets by smoothed local overdensity, we detect acoustic peaks in the six separate autocorrelation and cross-correlation functions of the sub-samples. Fitting models to these correlation functions, calibrated by mock galaxy and dark matter catalogues, we find that the inferred distance scale is independent of environment, and consistent with the result of analysing the combined sample. The shape of the baryon acoustic feature, and the accuracy of density-field reconstruction in the Zeldovich approximation, varies with environment. By up-weighting underdense regions and down-weighting overdense regions in their contribution to the full-sample correlation function, by up to |$50{{\ \rm per\ cent}}$|⁠, we achieve a fractional improvement of a few per cent in the precision of baryon acoustic oscillation fits to the CMASS data and mock catalogues: the scatter in the preferred-scale fits to the ensemble of mocks improves from |$1.45{{\ \rm per\ cent}}$| to |$1.34{{\ \rm per\ cent}}$| (pre-reconstruction) and |$1.03{{\ \rm per\ cent}}$| to |$1.00{{\ \rm per\ cent}}$| (post-reconstruction). These results are consistent with the notion that the acoustic peak is sharper in underdense environments.