Objective: Quantify effect of alternative econometric models, in estimating total direct medical expenditure in diabetes, arthritis, cardiac diseases, asthma, hypertension, stroke, and emphysema. Methods: Data from the MEPS's Household Component (2004), a nationally representative survey of the U.S. civilian noninstitutionalized population, was used. Accounting for the survey's clusters, strata and sampling weights; total direct medical expenditure was estimated under 11 different econometric models to assess their sensitivity to the zero mass, non-negative values and skewed distribution. Models compared were, OLS on raw-expenditure (OLSraw); OLS on log-expenditure (LnHom) and homoskedastic-retransformation; OLS on log-expenditure (LnHet) and heteroskedastic-retransformation; GLM with log-link and Gamma-family (GLMGam); GLM with log-link and Poisson-family (GLMPoi); Extended Estimating Equations (EEE); and 2-part models of OLSraw (2POLSraw); LnHom (2PLnHom); LnHet (2PLnHet); GLMGam (2PGLMGam); and GLMPoi (2PGLMPoi). Incremental expenditure from the method of recycled predictions summed over diseased population gave total expenditure. Box-Cox test confirmed log-link for GLM models while Modified Park's test determined a distribution between the Poisson and the Gamma for the family. LINK, RESET and Hosmer-Lemeshow test determined model fit, while COPAS test was employed for over-fitting and cross validation. Covariates included age, gender, race, ethnicity, marital status, education, insurance status, and comorbidity. Results: Total expenditure in billions (b) of dollars, for diabetes ranged from $48.5b(2POLSraw) to $127b(LnHet). Similarly, expenditure of arthritis ranged from $73b(2POLSraw) to $196b(LnHet); cardiac diseases ranged from $99.2b(2POLSraw) to $194b(LnHet); asthma ranged from $27.8b(2POLSraw) to $64.2b(GLMGam); emphysema ranged from $2.1b(2PLnHom) to $18.4b(LnHet); hypertension ranged from $69.9b(2POLSraw) to $241b(LnHet) and stroke ranged from $13.1b(2PLnHom) to $39.3b(LnHet). LnHet model was consistently associated with the highest total expenditure estimate, while 2POLSraw model typically predicted the lowest estimate. Conclusion: The strong influence of model choice on the total medical expenditure estimate, underscores importance of understanding the data generation procedure before selection of the appropriate estimator.