Purpose: To characterize and minimize the magnetization transfer (MT) effect in MR fingerprinting (MRF) relaxation measurements with a 2-pool (2P) MT model of multiple tissue types. Theory and Methods: Semisolid MT effect in MRF was modeled using 2P Bloch-McConnell equations. The combinations of MT parameters of multiple tissues (white [WM] and gray matter [GM]) were used to build the MRF dictionary. Both 1-pool (1P) and 2P models were simulated to characterize the dependence on MT. Relaxations measured using MRF with spin-echo saturation-recovery (SR) or inversion-recovery preparations were compared with conventional SR-prepared T 1 and multiple spinecho T 2 measurements. The simulations results were validated with phantoms and brain tissue samples. Results: The MRF signal was different from the 1P and 2P models. 1P MRF produced significantly (P < .05) underestimated T 1 in WM (20-30%) and GM (7-10%), while 2P MRF measured consistent T 1 and T 2 in both WM and GM with conventional measurements (pairwise test P > .1; correlated P < .05). Simulations showed that SRprepared MRF measuring T 1 had much less errors against the variation of the macromolecular fraction. Compared with inversion-recovery preparation, SR-prepared MRF produced higher relaxation correlations (R > 0.9) with conventional measurements in both WM and GM across samples, suggesting that SR-prepared MRF was less sensitive to the compositive effect of multiple MT parameters variations. Conclusions: 2P MRF using a combination of MT parameters for multiple tissue types can measure consistent relaxations with conventional methods. With the 2P models, SR-prepared MRF would provide an option for robust relaxation measurement under heterogeneous MT. K E Y W O R D S MRF, MT, saturation-recovery 728 | MENG Et al.