PURPOSE To assess the reproducibility of semiquantitative and quantitative analysis of first-pass myocardial perfusion cardiovascular magnetic resonance (CMR) in healthy volunteers. MATERIALS AND METHODS Eleven volunteers underwent myocardial perfusion CMR during adenosine stress and rest on 2 separate days. Perfusion data were acquired in a single mid-ventricular section in two cardiac phases to permit cardiac phase reproducibility comparisons. Semiquantitative analysis was performed to derive normalized upslopes of myocardial signal intensity profiles (myocardial perfusion index, MPI). The quantitative analysis estimated absolute myocardial blood flow (MBF) using Fermi-constrained deconvolution. The perfusion reserve index was calculated by dividing stress by rest data. Two observers performed all the measurements independently. One observer repeated all first scan measurements 4 weeks later. RESULTS The reproducibility of perfusion CMR was highest for semiquantitative analysis with an intraobserver coefficient of variability (CoV) of 3%-7% and interobserver CoV of 4%-10%. Semiquantitative interstudy comparison was less reproducible (CoV of 13%-27%). Quantitative intraobserver CoV of 10%-18%, interobserver CoV of 8%-15% and interstudy CoV of 20%-41%. Reproducibility of systolic and diastolic phases and the endocardial and epicardial myocardial layer showed similar reproducibility on both semiquantitative and quantitative analysis. CONCLUSION The reproducibility of CMR myocardial perfusion estimates is good, but varies between intraobserver, interobserver, and interstudy comparisons. In this study semiquantitative analysis was more reproducible than quantitative analysis.