Magnetic Resonance Fingerprinting (MRF) is a promising technique for quantitative multi-parametric measurements and tissue characterization [1, 2]. A train of RF pulses produces an MR signal that serves as a fingerprint for a certain tissue type. This fingerprint can later be compared with a dictionary of known signals. Many previously reported approaches combine a long fingerprint sequence with spiral or radial sampling for fast acquisition. However, these sequences may not be available on general-purpose MRI systems. Therefore, it would be desirable to also use Cartesian sampling in combination with parallel imaging for accelerated acquisition. In such a case, the MRF pulse train must be applied several times to cover k-space. In order to make all MRF signals start with a well-defined spin state, one possible approach is to insert a long delay time between the MRF pulse trains to allow the spin system to relax to its equilibrium state, which can be very time-consuming. In this work, we present a steady-state MRF approach, where the delay between MRF pulse trains can be very short and a stationary solution for the fingerprint signal is calculated. In this way, Cartesian or otherwise densely sampled MRF scans can be accelerated significantly without compromising matching accuracy. The findings are in line with earlier observations for accelerating inversion recovery experiments [4].The fingerprint sequence used in this study is based on a spoiled gradient-echo sequence [2]. It consists of a train of 200 flip angles, preceded by an inversion pulse. The repetition times (TR) are constant at 15ms to minimize B0 artifacts, leading to a total time of 3 seconds for the 200 steps. For dictionary generation, the sequence is repeated several times, and the expected signal of the complete pulse train is calculated using an Extended Phase Graph formalism [3] (Figure 1). The fingerprint signal reaches a stationary shape. Measurements have been performed on a Philips Achieva 1.5T system using an 8-channel head coil with a SENSE factor of 4 and a phantom equipped with known gel samples (Diagnostic Sonar, Eurospin II). The MRF dictionary consists of 18 entries for the different samples (227ms ≤ T₁ ≤ 1646ms; 48ms ≤ T₂ ≤ 369ms), 12 of which are present in the phantom, and one entry for the background.When the delay between the repetitions is short, the spin system will not have relaxed completely before the next inversion pulse. While the signal responses of the short-T₁ system (Figure 1, middle) are almost identical for each repetition, the long-T₁ system (Figure 1, bottom) exhibits different signal responses. However, after three repetitions, even this system reaches a stationary fingerprint. In the first measurement, the delay between the MRF sequence repetitions is large (5s), so that a fully relaxed spin system can be assumed each time. Using a dictionary calculated with this assumption, all 12 sample numbers are matched correctly (Figure 2, left). When the delay is reduced to 0.5s, the same dictionary leads to wrong matches, because the assumption of a relaxed spin system is violated (Figure 2, middle). When calculating a train of three sequence repetitions and using the steady-state fingerprint signal for the dictionary, all samples are matched correctly again (Figure 2, right). The first measurement took 4:17 min, the second one only 1:53 min.Quick repetition of MR fingerprinting sequences leads to stationary signal responses. With dictionaries based on these steady-state fingerprints, acquisition can be accelerated significantly. Using our method, we could accelerate a Cartesian SENSE measurement by more than a factor of two without reducing the match accuracy. The method is also applicable for segmented radial or spiral sampling. However, shortening the MRF repetition period (and thus the distance of the inversion pulses) may also lead to decreased differentiation of long T₁ values, so that a tradeoff is necessary for each application.