Extreme heat is a leading cause of weather-related mortality. Most research has considered the aggregate response of the populations of large metropolitan areas, but the focus of heat-related mortality and morbidity investigations is shifting towards a more fine-scale approach in which impacts are measured in smaller units such as postal codes. However, most existing statistical techniques to model the relationship between temperature and mortality cannot be directly applied to the intra-city scale because small sample sizes inhibit proper modelling of seasonality and long-term trends. Here we propose a time series technique based on local-scale mortality observations that can provide more reliable information about vulnerability within metropolitan areas. The method combines a generalised additive model with direct standardisation to account for changing death rates in intra-city zones. We apply the method to a 26-year time series of postal code-referenced mortality data from Philadelphia County, USA, where we find that heat-related mortality is unevenly spatially distributed. Fifteen of 46 postal codes are associated with significantly increased mortality on extreme heat days, most of which are located in the central and western portions of the county. In some cases the local death rate is more than double the county average. Identification of high-risk areas can enable targeted public health intervention and mitigation strategies.