Brain-machine interfaces (BMI) are powerful devices for restoring function to people living with paralysis. Leveraging significant advances in neurorecording technology, computational power, and understanding of the underlying neural signals, BMI have enabled severely paralyzed patients to control external devices, such as computers and robotic limbs. However, high-performance BMI currently require highly invasive recording techniques, and are thus only available to niche populations. Here, we show that a minimally invasive neuroimaging approach based on functional ultrasound (fUS) imaging can be used to detect and decode movement intention signals usable for BMI. We trained non-human primates to perform memory-guided movements while using epidural fUS imaging to record changes in cerebral blood volume from the posterior parietal cortex -a brain area important for spatial perception, multisensory integration, and movement planning. Using hemodynamic signals acquired during movement planning, we classified left-cued vs. right-cued movements, establishing the feasibility of ultrasonic BMI. These results demonstrate the ability of fUS-based neural interfaces to take advantage of the excellent spatiotemporal resolution, sensitivity, and field of view of ultrasound without breaching the dura or physically penetrating brain tissue.