Like many other places in the world, Hong Kong has drink-driving legislation which prohibits a driver from having in his blood alcohol exceeding a prescribed limit while in control of a motor vehicle. The accuracy of measuring this alcohol concentration is obviously of prime concern as an erroneous result can avert the administration of justice. The common practice is to deduct all errors from the measured value and compare the deducted value with the prescribed limit, so that the benefit of all errors of the measurement is given to the driver. It is therefore important for any laboratory responsible for measuring blood alcohol concentrations to identify and quantify all errors associated with the measurement. The present study examined 900 blood alcohol determinations carried out by the Hong Kong Government Laboratory (HKGL) on cases of suspected drink driving. The determinations were performed by 5 different analysts with two different sets of instruments during 1995-1997. Statistical analysis indicated that the instruments had no bearing on the random error or variability and that even though analyst was a significant factor on variability, the deviation from the mean so caused was only 0.3% and of no practical significance. When the systematic error introduced by the tolerance limits of the certified alcohol standards (purchased from the Laboratory of Government Chemists, UK) was taken into account, the total uncertainty (random plus systematic errors) of an alcohol determination at 99.5% confidence level was found to be 4%. It is recommended that laboratories engaged in blood alcohol determination should adopt similar statistical treatment of their analytical results to find out the error and to ensure that the results are independent of analyst and instrument used.