A technique for focusing ultrasound through the human skull is described and verified. The approach is based on a layered wavevector-frequency domain model, which propagates ultrasound from a hemisphere-shaped transducer through the skull using input from CT scans of the head. The algorithm calculates the driving phase of each transducer element in order to maximize the signal at the intended focus. This approach is tested on ten ex vivo human skulls using a 0.74 MHz, 320-element array. A stereotaxic reference frame is affixed to the skulls in order to provide accurate registration between the CT images and the transducer. The focal quality is assessed with a hydrophone placed inside the skull. In each trial the phase correction algorithm successfully restored the focus inside the skull at a location within 1 mm from the intended focal point. The results demonstrate the feasibility of using the method for completely non-invasive ultrasound brain surgery and therapy.