To assess the reproducibility of Harmonic Phase (HARP) analysis of myocardial MR tagged images acquired in the Multi-Center Study of Atherosclerosis (MESA). Using the HARP method, three independent observers performed two separate quantitative strain analyses of myocardial cine MR-tagging images blindly in 24 participants. The images were obtained in four different centers and analyzed at a single core lab. Each study comprised 3 short-axis slices subdivided in 12 segments (24 x 3 x 12 = 864 segments), each with three layers. Normal strains (circumferential [Ecc] and radial [Err]), principal strains (Lambda1, Lambda2), and the angle alpha (between Ecc-Lambda2) were calculated. Intraclass correlation coefficient (R) for peak systolic strains, and all pooled systolic and diastolic strain data were used to determine inter- and intraobserver agreement. Two observers also visually graded study quality. R values were related to the image quality in different myocardial regions and layers. Overall, HARP yielded an excellent inter- and intraobserver agreement for peak systolic strain data (for Ecc, R = 0.84 and 0.89, respectively) and all systolic pooled data (for Ecc, interobserver R = 0.82, intraobserver R = 0.69-0.76). Both inter and intraobserver agreement were lower for diastolic pooled data (R = 0.69 and 0.58-0.62, respectively). There was a direct relationship between image quality and performance of the HARP analysis, with increasing inter- and intraobserver R values in studies with longer tag persistence. Both inter- and intraobserver agreement were better in the anterior and septal myocardial regions, and in the midwall layer. The intraobserver agreement was similar among the three observers. Employing the HARP method for quantitative strain analysis of myocardial MR tagged images provides a high inter- and intraobserver agreement. These good results are obtained in case of good to excellent MR image quality.