R2* and susceptibility may be used to as means to probe myelin and iron in white matter and grey matter [1], which are important biomarker for studying the brain’s development and neurological disorders. Multi-center studies are often advantageous for involving participant of varying pathological and geological conditions, and becoming popular in neurological studies. Hence stability of the obtained quantitative R2* and susceptibility is a key issue for their widespread applications, given the many factors that may introduce variance to the final measurements. Previous studies [2,3] have involved only limited number of scanners from different vendors and coils, and reported varying level of variance of the two measures. In this work, the stability of simultaneously obtained R2* and susceptibility are investigated and compared across 10 sites that are equipped with the same scanner and receiver coil, and also the same processing software was used to achieve consistency of experimental setup to test the stability.Two healthy adult volunteers (one male and female, age 31 and 29) participated in this study. Multi-echo 3D SPGR acquisitions were performed at 10 different sites installed with a MR 750 3.0 T scanner (GE, Milwaukee) equipped with an 8-channel head coil. 12 echoes were acquired with a final TE of 29.7ms and TR of 35ms. 1mm isotropic whole brain coverage was achieved. All the acquired complex images were first registered using FLIRT software based on the magnitude image from the 10th echo. Then various ROIs (Fig.1) were defined on the susceptibility images and then transferred to the R2* images. Average values within the ROIs were taken as the measurements. R2* maps were obtained using voxel wise mono-exponential fitting, whereas susceptibility maps were derived using the STI suite toolbox available online.In order to better visualize the level of variance of R2* and susceptibility across different sites, all the measurements in different ROIs were first normalized by their median value, and then the box-whisker plots normalized measurements are made (Fig.2a). In this way, different measurements from different ROIs are set the same level, so their comparative variance may be better visualized. Firstly, it can be seen that the variance of all measurements from different sites, as indicated by the extent of the two whiskers, fall within $$$\pm$$$20% from the median value, including two outliers that are outside 99.3% data percentile as indicated by the red markers; in many cases, the variances of the measurements were constrained within $$$\pm$$$10% from the median value. Secondly, it is seen that the stability of R2* and susceptibility varies across different ROIs and subjects. Thirdly, it was observed that R2* showed a better level of stability as compared to the susceptibility, as indicated by the extents of the whiskers. In Fig.2b, strong positive linear correlations were observed between susceptibility and R2*, and visually the resulting linear fitting were in great consistency among measurements from different sites. The variations of the fitting were assessed by the mean and standard derivations of the slopes and intercepts of the resulting linear fitting as shown in the top left corner of the plots. The level of variance of the fitted linear model is in the similar range to those of the direct measurements, and similar level of stability of the fitted linear model is observed in the two subjects.In-vivo stability of R2* and susceptibility at 3T were great: for susceptibility measure, the largest variations of measurements in all the chosen ROIs fall within $\pm$20%, which translate to about $$$\pm$$$20ppb for GP and as little as for $$$\pm$$$4ppb CN, this is considerably smaller than the previously reported variations as blessed by the more consistent experiment setup; the stability of R2* was slightly better due to its deterministic computation. Also similar level of stability is achieved in the linear fitting between R2* and susceptibility. Given all the controllable factors that may affect the susceptibility measurements have been accounted for, other inevitable sources of error include noise, registration error, field inhomogeneity, subject orientation and computation bias may contribute to the measurement variations.