Magnetic resonance fingerprinting (MRF) uses a random set of flip angles (FA) and repetition times (TR) to simultaneously sensitize a scan to multiple tissue properties [1]. In previous work we described an optimization method that allows significant reduction (2 orders of magnitude) in the required number of acquisitions [2]. In this work we demonstrate an application of our method using a Cartesian, fully sampled, gradient-echo planar imaging sequence to acquire multiple tissue maps from a single slice on a clinical 1.5 T scanner in 2.4 seconds. To our knowledge, this represents the fastest scan time yet achieved by an MRF acquisition. We validate the proposed method in a phantom and demonstrate the utility of the method in a healthy human volunteer. A FA/TR schedule of length N=16 was generated using the optimization method previously described [2]. The EPI sequence was modified, as shown in Figure 1, with the excitation α_i and delays TR⁽ⁱ⁾ set according to the optimized schedule. All experiments were conducted on a 1.5T Siemens Avanto scanner (Siemens Healthcare, Erlangen, Germany) using a 32-channel head coil. A phantom composed of vials with varying concentrations of doped water was used to mimic the different tissue parameters in the human brain. The phantom was scanned with the following acquisition parameters: FOV=240×240 mm², Slice thickness=5 mm, matrix size=128×128 and receiver bandwidth=1536 Hz/pixel. Total scan time was 2.4 seconds for the N=16 schedule used. The T₁/T₂ maps obtained were validated by comparison with a spin-echo sequence with varying TE (=6,12,24,36,72,144 and 200ms) and TR (=50, 100,200,400,600,800,1000,1200,1600,2000,2400 and 3200ms) times where the T₁/T₂ maps were extracted by a 3-point fitting of the acquired data to the signal equation. To measure the agreement between the two sequences, the coefficient of determination (R²) was calculated. For the in vivo experiments a healthy 31 year old male subject was recruited and provided informed IRB-approved consent. The subject was scanned using the same acquisition parameters as the phantom. The tissue maps were reconstructed by matching to a pre-computed dictionary as with standard MRF [1]. The proton density, T₁,T₂ and B₀ maps for the phantom are shown in Figure 2a-d. The T₁/T₂ values obtained with the proposed sequence versus the spin echo sequence are shown in Figure 3. The computed R² for the T₁/T₂ maps was 0.96/0.90, indicating strong correlation between the two acquisition methods. Tissue maps for the healthy volunteer are shown in Figure 3.Contrary to other MRF sequences that heavily undersample (up to 1/48th) k-space, in this sequence a full Cartesian k-space sampling was used. This feature of the proposed sequence is highly beneficial in that: i) reconstruction is simplified (simple Fourier transform rather than regridding), ii) undersampling artifacts are avoided and iii) further acceleration by undersampling is readily achievable using parallel imaging and and/or simultaneous multi-slice methods [3]. A limitation of the proposed sequence, however, is its sensitivity to field inhomogeneities that is common to EPI which requires care when porting this sequence to higher fields. Nevertheless, since EPI based sequences have been successfully used at high fields (7T) [4] this sequence is likely to work as well.We have demonstrated and validated an in vivo application of our optimized MRF sequence on a clinical scanner. The scan time achieved by this method is, to our knowledge, the fastest yet obtained for an MRF sequence. Potential applications of this method include clinical imaging of pediatric patients and imaging of dynamic processes such as dynamic contrast-enhanced MRI.