Abstract Duplicate breath alcohol testing from each individual provides confidence in the results when reasonable agreement (i.e. ± 0.02 g/210 L) is achieved. For this reason many jurisdictions require duplicate testing. The State of Washington has recently implemented an infrared breath testing program and now requires two breath samples from each individual. Statistical analysis of 1847 duplicate breath tests is presented. Three variables are analyzed: first alcohol result (ALC1), the absolute difference between the two breath samples (DIFFA), and the signed difference between the two breath samples (DIFFS). The first breath alcohol result ranged from 0.021 to 0.338 g 210 L with a mean of 0.157 g/210 L. The absolute difference ranged from 0.00 to 0.05 g 210 L . The signed difference ranged from −0.05 g 210 L to 0.05 g 210 L . The absolute difference was regressed upon the first alcohol result and resulted in poor linear correlation of r = 0.212. Duplicate breath test differences do not appear to be a function of subject's alcohol level, but rather of sample provision.