We explore a methodological improvement to the standard dynamic demand model for petrol - a general model which allows for slowly evolving, unobservable habits. If this habit formation model is correct, then standard estimation techniques produce inconsistent estimates. We find price elasticities of -0.13 (short-run) and -0.20 (long-run). Importantly, standard techniques are misleading about the precision of elasticity estimates and the confidence interval around the long-run price elasticity is quite wide. We test for price irreversibility and find, in contrast to the USA, almost no evidence that petrol responds differently to price increases and decreases.