BACKGROUND: Accurate recording of ethnicity in electronic healthcare records is important for the monitoring of health inequalities. Yet until the late 1990s, ethnicity information was absent from more than half of records of patients who received inpatient care in England. In this study, we report on the usefulness of a names-based ethnicity classification, Ethnicity Estimator (EE), for addressing this gap in the hospital records. MATERIALS AND METHODS: Data on inpatient hospital admissions were obtained from Hospital Episode Statistics (HES) between April 1999 and March 2014. The data were enhanced with ethnicity coding of participants' surnames using the EE software. Only data on the first episode for each patient each year were included. RESULTS: A total of 111,231,653 patient-years were recorded between April 1999 and March 2014. The completeness of ethnicity records improved from 59.5 % in 1999 to 90.5 % in 2013 (financial year). Biggest improvement was seen in the White British group, which increased from 55.4 % in 1999 to 73.9 % in 2013. The correct prediction of NHS-reported ethnicity varied by ethnic group (2013 figures): White British (89.8 %), Pakistani (81.7 %), Indian (74.6 %), Chinese (72.9 %), Bangladeshi (63.4 %), Black African (57.3 %), White Other (50.5 %), White Irish (45.0 %). For other ethnic groups the prediction success was low to none. Prediction success was above 70 % in most areas outside London but fell below 40 % in parts of London. CONCLUSION: Studies of ethnic inequalities in hospital inpatient care in England are limited by incomplete data on patient ethnicity collected in the 1990s and 2000s. The prediction success of a names-based ethnicity classification tool has been quantified in HES for the first time and the results can be used to inform decisions around the optimal analysis of ethnic groups using this data source.