BACKGROUND: Caffeine intake from one cup of coffee one hour before adenosine stress tests, corresponding to serum caffeine levels of 3-4 mg/L, is thought to be acceptable for non-invasive imaging. AIMS: We aimed to elucidate whether serum caffeine is independently associated with adenosine-induced fractional flow reserve (FFR) overestimation and their concentration-response relationship. METHODS: FFR was measured using adenosine (FFRADN) and papaverine (FFRPAP) in 209 patients. FFRADN overestimation was defined as FFRADN - FFRPAP. The locally weighted scatterplot smoothing (LOWESS) approach was applied to evaluate the relationship between serum caffeine level and FFRADN overestimation. Multiple regression analysis was used to determine independent factors associated with FFRADN overestimation. RESULTS: Caffeine was ingested at <12 hours in 85 patients, at 12-24 hours in 35 patients, and at >24 hours in 89 patients. Multiple regression analysis identified serum caffeine level as the strongest factor associated with FFRADN overestimation (p<0.001). The LOWESS curve demonstrated that FFRADN overestimation started from just above the lower detection limit of serum caffeine and increased approximately 0.01 FFR unit per 1 mg/L increase in serum caffeine level with a linear relationship. The 90th percentile of serum caffeine levels for the ≤12-hour, the 12-24-hour, and the >24-hour groups corresponded to FFRADN overestimations by 0.06, 0.03, and 0.02, respectively. CONCLUSIONS: Serum caffeine overestimates FFRADN values in a linear concentration-response manner. FFRADN overestimation occurs at much lower serum caffeine levels than those that were previously believed. Our results highlight that standardised caffeine control is required for reliable adenosine-induced FFR measurements.