This paper deals with building bootstrap tests for comparing the mean costs between two groups of patients undergoing different health care strategies in the context of strongly skewed and leptokurtic data. In medico-economic evaluations, the distribution of data is frequently skewed and leptokurtic, because a few patients can produce large costs. Consequently, traditional methods for comparing the mean costs between two treatments are inappropriate. In this paper, nonparametric bootstrap procedures proposed or suggested in the literature are analysed and improved. Unfortunately, despite these improvements, these nonparametric methods fail in case of too small sample size, because of their inability to take into account the probabilities in the statistic distribution tails. To solve this problem, we develop a parametric bootstrap method. Lastly, Monte Carlo experiments are carried out, using resampling from real data, for assessing the various tests performance