We showed that three shells are sufficient to result in good approximations of MAP-MRI indices from numerical simulation. We used multiple compartment microstructure models to fit the two shell data and extrapolate the third shell with a higher b-value. We compared the performance of two models, NODDI and NODDI with fiber crossing (NODDIx), on the Human Connectome Project (HCP) DWI data. NODDIx showed improvement in the white matter with extrapolation but NODDI did not. Both NODDI and NODDIx failed to improve the results in the gray matter. Our approach also provides a new mechanism in validating or comparing microstructure models.