OBJECTIVES: To apply bootstrap techniques to derive confidence intervals (CIs) for cost differences between two anaesthetic regimes, and to compare with standard parametric methods for deriving CIs. METHODS: In a cardiac surgery clinical trial, patients were randomized to one of two anaesthetic regimes, either propofol or midazolam, for the induction and maintenance of anesthesia, and sedation post-surgery in ICU. A cost comparison of these two regimes was published in 1996, with all drug usage and ICU nursing time costs. Parametric methods were used to construct 95% CIs for the difference between propofol and midazolam. Untransformed parametric analysis of cost data often fails because of distributional skewness. Therefore, log-transformed analysis was undertaken to overcome the skewness problem. More recently, log-transformation has become a discredited strategy because healthcare decision-makers focus on total budgets – for instance, the total annual budget available to provide surgery at a particular centre. The only estimator directly linked to this total cost is the arithmetic mean – therefore any analysis based upon geometric means, arising from log-transformation, is inappropriate. Other authors have therefore recommended bootstrap techniques to produce CIs for cost differences on an untransformed scale. This approach is not invalidated by distributional skewness. RESULTS: With 37 propofol and 33 midazolam patients, a comparison of drug plus nursing costs showed an advantage per subject for propofol of £43.23 (95% CI: -£3.21 to £89.67). An untransformed two-sample t-test proved inappropriate, because the necessary distributional assumptions were not satisfied. Bootstrap CIs for the cost difference were then constructed using four different bootstrap techniques. All bootstrap CIs showed a striking similarity to the untransformed estimates above, demonstrating the robustness of the t-test and conventional parametric methods, despite distributional skewness. CONCLUSIONS: Both untransformed and bootstrap estimates should be presented – hopefully, the two methods will be in agreement and provide mutually supportive evidence.