Aims: Large discrepancies are typically found between per capita alcohol consumption estimated via survey data compared with sales, excise or production figures. This may lead to significant inaccuracies when calculating levels of alcohol-attributable harms. Using British data, we demonstrate an approach to adjusting survey data to give more accurate estimates of per capita alcohol consumption. Methods: First, sales and survey data are adjusted to account for potential biases (e.g. self-pouring, undersampled populations) using evidence from external data sources. Secondly, survey and sales data are aligned using different implementations of Rehm et al.'s method [in (2010) Statistical modeling of volume of alcohol exposure for epidemiological studies of population health: the US example. Pop Health Metrics 8, 1-12]. Thirdly, the impact of our approaches is tested by using our revised survey dataset to calculate alcohol-attributable fractions (AAFs) for oral and pharyngeal cancers. Results: British sales data under-estimate per capita consumption by 8%, primarily due to illicit alcohol. Adjustments to survey data increase per capita consumption estimates by 35%, primarily due to under-sampling of dependent drinkers and under-estimation of home-poured spirits volumes. Before aligning sales and survey data, the revised survey estimate remains 22% lower than the revised sales estimate. Revised AAFs for oral and pharyngeal cancers are substantially larger with our preferred method for aligning data sources, yielding increases in an AAF from the original survey dataset of 0.47-0.60 (males) and 0.28-0.35 (females). Conclusion: It is possible to use external data sources to adjust survey data to reduce the under-estimation of alcohol consumption and then account for residual under-estimation using a statistical calibration technique. These revisions lead to markedly higher estimated levels of alcohol-attributable harm.