This analysis compares the difference between the Energy Performance Certificate (EPC)-modelled and smart meter measured annual energy use on a like-for-like basis in 1,374 gas-heated British households selected from the Smart Energy Research Lab (SERL) Observatory. EPCs and metered energy use were converted to primary energy use intensity to provide a comparison of the same quantity for the first time.We show that EPCs predict substantially and significantly more energy use than metered in homes in Great Britain. EPC bands A and B show no statistically significant difference, but all other bands show a statistically significant and increasing gap as EPC rating decreases. Average energy use in low efficiency bands is very similar, despite large differences in EPC-modelled energy use. This difference persists after removing homes matching the EPC model assumptions regarding occupancy, thermostat set-point and whole-home heating; suggesting that differences in modelled and actual behaviour are unlikely to fully explain the discrepancy. EPCs are a core tool in the residential energy sector, and the gap between EPC-modelled and metered energy use could have a significant impact on policy, research and industry. Future research should investigate this discrepancy and its implications, and action should be taken to improve EPCs.