The mean spherical technique (SMT) has been proposed as an attempt to disentangle microscopic diffusion properties from the mesoscopic tissue organization. However, it remains unclear under which conditions exactly the technique could still fruitfully deliver its metrics. In this study, SMT’s microscopic fractional anisotropy estimates (μFA) are investigated using synthetic and high-quality diffusion data. We find that any compartmental heterogeneity can crucially impact the µFA extracted from SMT. In addition, the b-value ranges where SMT delivers accurate information is specific to the microstructure itself. Our work suggests that SMT-driven µFA should be examined with care.