Statistical power in neuroscience studies is often limited, leading to, among others, overestimates of effect size in studies with low subject numbers, low reproducibility of results, and a reduced likelihood that a statistically significant result reflects a true effect [1]. However, to assess the power of a given study and sample size, one needs to have an estimate of the effect size. In diffusion MRI (dMRI) analysis, estimating the effect size is made difficult by two challenges. First, it is often not feasible to scan many subjects, essential for a reasonable effect size estimate. Here, we use 100 subjects of the 3T three-shell HARDI data from the HCP WU-Minn Consortium [2] to address the sample size requirements. Second, dMRI image analysis can require substantial computing time, especially when using parametric biophysical compartment dMRI models like CHARMED [3] and NODDI [4], that can have higher specificity than diffusion tensor imaging (DTI). Here, we use the flexible, modular GPU-accelerated Maastricht Diffusion Toolkit (MDT; [8]) to overcome prohibitive computing times. Using these tools, the aim of this study is to compute robust effect size estimates and assess variability and correlation of multiple dMRI microstructure modeling parameters focusing on the well-known microstructural architecture of the splenium, body and genu of the Corpus Callosum (CC).The 100 non-related subjects 3T dataset (1.25mm isotropic, three shells of b=1000, 2000, 3000 s/mm^2, 90 directions each and 18 b0 volumes) from the HCP WU-Minn Consortium [2] were used for analysis in their preprocessed form [5]. Each dataset was fitted with the following models using MDT [8]: Tensor [6], Ball&Stick [7], NODDI [4] and CHARMED [3]. The Ball&Stick model was initialized using an S0 fit; all the other models were initialized using the theta and phi angles of the Ball&Stick fit. Average runtime of final model optimization per subject using a R9 280x ATI graphics card was 00h:02m:10s (non-linear DTI), 00h:01m:30s (Ball&Stick), 00h:06m:30s (NODDI), 00h:49m:30s (CHARMED) and a total runtime of approximately 134 hours over all subjects. A white matter skeleton was created from the average DTI FA map in MNI-152 space using the standard TBSS [9] pipeline and the following parameter maps were projected to this skeleton: FA (Tensor), SF (Stick Fraction; Ball&Stick), FR (Fraction of Restricted compartment; CHARMED), NDI (Neurite Density Index; NODDI) and FI (Fraction of Intracellular compartment; NODDI). Next, we aligned the JHU-ICBM-FA Mori [10] white matter atlas to the skeleton and applied the skeleton mask to arrive at a white matter parcellation map. We then extracted, for each subject and for each projected parameter map, the mean over voxel values for three corpus callosum regions of interests (ROIs): the splenium (3351 voxels), body (3379 voxels) and genu (2204 voxels). We computed mean voxel value over each of these ROIs, for each subject and model. The ROI and parameter specific group mean, standard deviation (StDv) and parameter correlations are reported over the 100 per-subject ROI mean values. Finally, to estimate effect sizes for CC splenium-body-genu differences, the Cohen's d factor was computed between the splenium-body and genu-body as: $$$\frac{\mu_1 - \mu_2}{\sqrt{(\sigma_1^2 + \sigma_2^2)/2}}$$$.Figure 1 shows that effect sizes range from low (d=0.022 for FA genu-body) to very high (d=3.4 for FR splenium-body). Large effect size of 1-3 can be observed for many of the differences, most consistently (for both genu-body and splenium-body) for NDI and WIC (these are simply related: NDI = WIC / (WIC + WEC)). Here it should be noted that high values for normalized parameters approaching 1 could compress StDv estimates and thus inflate effect size estimates. However, although some values are high (e.g. FR) we observed little such ceiling effects (see Figure 2). FS, FR and FA showed a strong effect for splenium-body but small effect for genu-body. In figure 2, constant value differences (intercepts) can be observed with FS<FA<FR~=NDI. One clear outlier can be observed (NDI~=1). High between-model correlations (slope) can be observed between FA/FS, and the triangle FR/WIC/FA, for all three ROIs approximately equally.Although differences between models are present and intra-subject CC differences are large, medium to large effect sizes (Cohen’s d=1-3) are observed for advanced dMRI microstructure measures. This is promising for power and reproducibility of dMRI microstructure studies.