AUDIENCE: Scientists interested in quantitative imaging, parameter mapping, and estimating/resolving partial volume effects. PURPOSE: A common problem in all types of imaging is the so-called partial volume effect that results from voxels with mixed content. Several investigators have previously used multi-exponential fitting to derive separate components of a voxel [1-3]. However, these methods are unstable and extremely sensitive to noise. A major cause of this problem is that the signal evolutions for the different components of the voxel are similar (i.e. exponential decays) and thus difficult to separate. A recently proposed framework, Magnetic Resonance Fingerprinting (MRF) [4,5], has the potential to open up numerous new possibilities for MR. Unlike the vast majority of MR acquisitions in which a fixed pulse sequence is repeated multiple times to encode spectral, spatial or decay information (or some combination of these quantities), MRF uses a pseudorandomized sequence which is simultaneously sensitive to multiple parameters during a single acquisition. This provides a rich signal that is no longer described by a simple exponential decay. Thus in MRF, signal evolutions can look very different and are more easily separable. We seek to demonstrate that MRF can resolve multiple material components from single voxels made up of several tissue types, validate the derived tissue fractions in a realistic simulation model and demonstrate its use in vivo. MATERIALS & METHODS: The fingerprint of each mixed voxel is modeled as a weighted sum of components of interest, D, whose signal evolutions are known a priori from the Bloch equations: Svoxel=Dw. The relative fraction attributed to the tissue type in each voxel is determined by solving for the weights using conventional pseudoinverse methods, namely (DD)DSvoxel=w. To this end, we investigate simulated MRF signals in heterogeneous mixed tissue. Two tissue types are considered: Tissue A, with T1=180ms and T2=75ms, and Tissue B with T1=1850ms and T2=150ms. A QUEST MRF sequence [5] was modeled to collect signal evolutions at 225 time points. A 225 image series (resolution