OBJECTIVES: To investigate the predictors of self-rated health status for the U.S. civilian noninstitutionalized population. METHODS: Cross-sectional analyses using multivariate logistic regression were performed with 11,405 individuals from the 2003 MEPS. The self-rated health status was dichotomized into two categories (fair/poor health and good/very good/excellent health). Standard demographic variables were employed as predictors in the regression, and the impact of a given predictor on the self-rated health status was obtained as odds ratios. To check the presence of multicollinearity, the conditional index and variance composition were examined. RESULTS: Smaller family size, very low incomes (less than $15,000/year), non-white ethnicity (Hispanic, African-American, and Asian), older ages, female, lower education than college, having public health insurance, and having worse mental health status were significantly and consistently associated with fair/poor perceived health (p<0.05). An additional family member and male gender were related with decreased risk of fair/poor perceived health by 0.88 and 0.87, respectively. Subjects with very low incomes, Hispanic ethnicity, and aged between 56 to 65 were associated with 1.19, 1.34, and 3.65 times more likely to rate their health as fair/poor in comparison with subjects with high incomes (more than $100,000/year), White ethnicity, and aged between 16 to 25, respectively. Similarly, persons with lower education (high school degree) and public insurance would be 1.77 and 1.75 times as likely as persons with Bachelor's degree and without insurance to rate their health as fair/poor, respectively. An additional score of mental health status (1 [excellent] ~ 5 [poor]) was associated with 2.67 times more likely to have fair/poor perceived health. The model did not suffer from significant multicollinearity since the maximum conditional index was less than 10. CONCLUSION: Health perceptions are strongly associated with standard demographic variables like income, ethnicity and age. Future work should address the causal factors behind these associations.