OBJECTIVES: The objective of this paper is to explain and illustrate the usefulness and limitations of bootstrapping methodologies and to improve applied health economics research by encouraging researchers to rely on rigorous empirical tests when selecting the most appropriate bootstrapping method. METHODS: Pair, parametric, nonparametric, and wild bootstrapping types are analyzed for linear, non linear, instrumental variable (IV) and discrete choice models. For each model, guidelines for selecting a consistent and efficient bootstrapping method are provided, and percentage deviations from the other methods are calculated. The Medstat MarketScan(r) Research Databases for 1995 - 2000 were used in this study. Patients with evidence of asthma were selected from claims, encounter, enrollment, and pharmaceutical data files. Separate models were estimated for patients in fee-for-service (FFS) and non-FFS plans based on likelihood tests. Total cost was estimated using linear, non linear, and IV approaches. The ratio of controller to reliever medication was used as an IV. Hospitalization was estimated using a discrete-choice model. Control variables included demographics, clinical, and county characteristics. RESULTS: We found that the selection of an inappropriate bootstrapping methodology can yield results that deviate greatly from the results produced using the appropriate methodology. Increase in bias in the estimation of standard errors affects the significance of the coefficients. CONCLUSIONS: There is no single bootstrapping method that can be applied in all situations since the behavior of the bootstrap depends critically on both the statistic being examined and approximation to the underlying population distribution function. In our example, the greatest deviance was attained between parametric residual bootstrapping and consistent and efficient wild bootstrapping when errors are heteroskedastic and not normal for estimation of non-linear model.