Multi-diffusion-time diffusion-weighted MRI can probe tissue microstructure, but the method has not been widely applied to the microvasculature. At long diffusion-times, blood flow in capillaries is in the diffusive regime, and signal attenuation is dependent on blood velocity ([Formula: see text]) and capillary segment length ([Formula: see text]). It is described by the pseudo-diffusion coefficient ([Formula: see text]) of intravoxel incoherent motion (IVIM). At shorter diffusion-times, blood flow is in the ballistic regime, and signal attenuation depends on [Formula: see text], and not [Formula: see text]. In theory, [Formula: see text] could be estimated using [Formula: see text] and [Formula: see text]. In this study, we compare the accuracy and repeatability of three approaches to estimating [Formula: see text], and therefore [Formula: see text]: the IVIM ballistic model, the velocity autocorrelation model, and the ballistic approximation to the velocity autocorrelation model. Twenty-nine rat datasets from two strains were acquired at 7 T, with [Formula: see text]-values between 0 and 1000 smm⁻² and diffusion times between 11.6 and 50 ms. Five rats were scanned twice to assess scan-rescan repeatability. Measurements of [Formula: see text] were validated using corrosion casting and micro-CT imaging. The ballistic approximation of the velocity autocorrelation model had lowest bias relative to corrosion cast estimates of [Formula: see text], and had highest repeatability.