<p>Quantification of the dopamine D_2/3 receptors tracer [¹⁸F]fallypride is usually performed by defining the cerebellum (CB) as the reference region for its use in kinetic modeling. In mouse studies, [¹⁸F]fallypride is defluorinated causing gradual uptake in the skull which, in addition to extra-striatal binding, can contaminate the CB activity, therefore introducing errors in reference region kinetic modeling. Here we propose a method using non-negative matrix factorization (NMF) to accurately extract the reference region time activity curve (TAC) unaffected by spill-over from surrounding regions. We compared the NMF method with template-based CB reference region (erodedCB), where the region was eroded to avoid spill-over. The different methods were applied in a drug challenge study using RX821002, and compared with results obtained using [¹¹C]raclopride. Striatal and brain parametric maps of nondisplaceable binding potential (<i>BP</i>_ND) were calculated with the different methods, and differences between baseline and challenge were investigated. The NMF reference region TACs showed higher peak and lower tail activity compared with erodedCB. Striatal <i>BP</i>_ND values calculated with NMF were about 20% higher compared to those calculated with erodedCB, and difference between baseline and challenge increased using NMF (NMF: 11.5%, erodedCB: 7.0%). Parametric t-statistic maps show clusters with significant differences using NMF but not with erodedCB. [¹¹C]raclopride <i>BP</i>_ND differences between baseline and challenge were lower (6.1%) than with [¹⁸F]fallypride, but <i>BP</i>_ND variability was lower. In summary, NMF allowed us to extract reference region TACs without contamination from skull or extra-striatal uptake, which improved voxel-wise detection of differences in a drug challenge study.</p>