Imaging the very short T2 tissues frequently encountered in the musculoskeletal system using MRI requires specialized pulse sequences with very short echo times (TE). To address this challenge, ultrashort TE (UTE) sequences such as 3DCones [1] typically begin data acquisition as soon as possible after the RF excitation and acquire k-space in a center out fashion [2]. Due to the fast decay rate of the transverse magnetization, diffusion weighting is challenging in short T2 tissues. Only very limited b-values can be achieved before the signal has decayed to zero. In this work, we used stimulated echoes to allow spin diffusion while the magnetization was stored along the longitudinal axis, therefore achieving useful b-values. We show that the technical challenges associated with this approach can be overcome and that useful diffusion weighting can be achieved in tissues with very short T2s.The pulse sequence is shown in Fig.1. The diffusion preparation module is repeated every TR period, and is immediately followed by N separate k-space spokes during short time intervals τ. The diffusion duration 𝛿 is kept sufficiently small (1-5ms) to minimize T2 decay, while the main evolution time 𝛥 is chosen large enough (30-100ms) to allow maximum b-values of 100 s/mm² or greater. Several challenges arise with this approach: The first challenge using stimulated echo diffusion preparation with small diffusion gradient areas is that the corresponding phase labeling oscillation wavelength in the diffusion direction can be larger than the pixel dimension and therefore appear as banding in the image (see Fig.2). The wavelength of these oscillations is given by: 𝜆=2π/(𝛾𝛿G). In typical diffusion sequences the value of 𝛿G is large and these oscillations do not artifact the image. The most straightforward way to mitigate this issue is to apply the diffusion gradients along the largest pixel direction which is typically the slice direction. The second challenge is scan time. To minimize the scan time, several k-space spokes can be acquired after each diffusion preparation. However this causes T1 recovery contamination (especially for the later spokes), so that the signals are artificially elevated for higher b-values, resulting in an underestimation of D. One possible solution to this problem is to reverse the phase of the second 90º pulse (see Fig.1) so that the magnetization is stored along the negative z-axis. This results in an underestimation of the signals due to T1 recovery. Finally, phase cycling the second 90º pulse and averaging the acquisitions helps to offset these effects. Bloch simulations were performed to study the signal behavior and quantitative accuracy of the diffusion measurements. Phantom scans were performed in a doped water phantom with T1/T2 = 100ms/65ms using a T/R head coil on a 3T GE HDxt clinical MR scanner. The first scans were performed with the diffusion gradient applied within the imaging plane to highlight the confounding oscillation patterns for low b-values. Next, multi b-value measurements were performed in the slice direction, with the diffusion gradient area large enough so that several phase labeling cycles are contained within a slice (𝜆 << slice) at b-values from 0-100 s/mm². Three separate acquisitions were performed, with the second 90º pulse chosen to either tip up the transverse magnetization, tip down the transverse magnetization, or cycle between both modes.Fig.2 shows two axial phantom images using different small diffusion gradient areas applied left/right. The left image shows the result with 𝜆 = 9mm, while the right one shows the result with 𝜆 = 46mm. Figs.3A-C shows the simulated signal decay curves as a function of b-value. Fig.3D shows the experimental results. The data shown in blue is acquired so that the diffusion prepared magnetization is stored along the positive z-axis before UTE imaging (resulting in higher signals), while the data in red was acquired so that it is stored along the negative z-axis (resulting in lower signals). Finally, the data shown in black is the cycled data which converges to the signal levels of the clinical EPI reference sequence (green). Stimulated echo based diffusion weighted 3DUTE MR imaging can be used effectively to achieve useful b-values in short T2 MSK tissues. Several technical challenges have to be addressed however, such as minimizing the oscillatory signal banding due to the low diffusion gradient area, and reducing T1 contamination in multi-spoke UTE acquisitions. This initial proof-of-principle study shows that this can be accomplished and that good quantitative agreement with clinical EPI diffusion sequences is possible.