OBJECTIVE: The association of caffeine intake with testosterone remains unclear. We evaluated the association of caffeine intake with serum testosterone among American men and determined whether this association varied by race/ethnicity and measurements of adiposity. METHODS: Data were analyzed for 2581 men (≥20 years old) who participated in the cycles of the NHANES 1999-2004 and 2011-2012, a cross-sectional study. Testosterone (ng/mL) was measured by immunoassay among men who participated in the morning examination session. We analyzed 24-h dietary recall data to estimate caffeine intake (mg/day). Multivariable weighted linear regression models were conducted. RESULTS: We identified no linear relationship between caffeine intake and testosterone levels in the total population, but there was a non-linear association (pnonlinearity < .01). Similarly, stratified analysis showed nonlinear associations among Mexican-American and Non-Hispanic White men (pnonlinearity ≤ .03 both) and only among men with waist circumference <102 cm and body mass index <25 kg/m2 (pnonlinearity < .01, both). CONCLUSION: No linear association was identified between levels of caffeine intake and testosterone in US men, but we observed a non-linear association, including among racial/ethnic groups and measurements of adiposity in this cross-sectional study. These associations are warranted to be investigated in larger prospective studies.