Super selective arterial spin labeling (ssASL) rely on mutual effects of phase accumulation by gradients and RF to selectively invert a spatial region, so that vessel-specific information such as perfusion [1] and MRA [2] can be obtained. The labeling profile determines the labeling efficiency, and hence consequently the robustness of the technique as selection of the labeling region is prone to operational errors and patient motions. The labeling profile should ideally be a binary ON/OFF state but is never the case in practice due to physics constrains. In this work, the gradient design scheme in the labeling train is investigated to improve the labeling profile, as compared to the currently adopted approach.The labeling strategy used for ssASL is illustrated in Fig.1: in-plane gradients are played out with RF in both the label and control acquisition. The effective gradient is rotated from pulse to pulse with accumulated phase added to RF, so that only the designated labeling region satisfies adiabatic condition. The implementation of the gradient could be either refocused gradient (RG) where the gradient is immediately negated resulting zero net phase (Fig.1a) or unrefocused gradient (UG) so that a net phase accumulation persists (Fig.1b). In conventional design, gradients in both label and control acquisition are refocused (RG/RG). We hypothesize that using UG in the control acquisition may lead to a flatter profile blessed by better immunity to the adiabatic pulse, which would consequently lead to more homogenous labeling profile that is the difference between label and control. To verify this hypothesis, numerical simulation based on Bloch equation was performed for spins (T1/T2: 1650/200 ms, velocity 10 to cm/s in interval of 10 cm/s) moving through tagging slab (50 mm) in presence of ssASL labeling region (Diameter: 20 mm), at an integration step of 2.5 us (Fig.2). Pulse sequence parameters are: Hanning pulse 500 us; tagging unit duration 1.184 ms; selective/average gradient 7/0.7 mT. In-plane selective gradient sequence was configured with rotation step of 18^o and amplitude modulation [3] implemented in period of 160 tagging units. In-plane profile resolution was 1mm and the overall labeling profile was obtained by averaging multiple layers of spin entering the system at variable time points in the labeling train. Different gradient design schemes were simulated (label/control): RG/RG, RG/UG, UG/RG, UG/UG, and the resulting profiles were compared. In vivo experiment was performed on a 3.0T whole body system (GE Discovery 750) using an 8 channel head coil. A volunteer was recruited and consent form was acquired prior to the study. Super selective ASL perfusion mapping of the right internal carotid artery (rICA) was performed with conventional R/R design and the proposed R/U design. Identical design of the ASL labeling module as in simulation was used with radius set to 50/20 mm in R-L/A-P direction, a 3D stack of spiral readout was used with the following parameters: FOV 200 mm, spiral number 4, NEX 2, post labeling delay 1525 ms. In order to evaluate the labeling profile, the center of the labeling region was intentionally shifted from 0 mm to 18 mm away from the center of the rICA in AP direction to observe the variation of the perfusion signal. The simulated labeling profiles of different labeling schemes are shown in Fig.3. It can be seen that the use of RG/UG gradient pair led to the best labeling profile within the selected radius. In the in-vivo result (Fig.4), comparing the resulting regional perfusion maps using RG/RG and RG/UG schemes, it can be seen that both schemes led to similar level of perfusion weighted signal when the labeling region was centered at the vessel, as the labeling center was increasingly moved away from the vessel, the RG/UG scheme offers consistently higher level of perfusion signal attributed to better labeling profile, hence improved level of robustness.Super selective ASL offers unique information that was previously only available in DSA, with non-ideal labeling profile (DSA may be considered to feature a binary labeling profile). In this work, the labeling gradient scheme was investigated and from simulation it was observed that the use of RG/UG gradient scheme features better labeling profile, as unrefocused gradient is beneficial for control acquisition. In-vivo experiment with various labeling offsets verified that this gradient scheme was more tolerant with the misalignment of the labeling center. Further improvement of the labeling profile may be feasible with redesigning of both the RF and gradient.