Directional beamforming has shown promising results for creating vector flow images. The method measures both the flow angle and the magnitude of the velocity. The flow angle is estimated by focusing lines in a range of angles from 0 to 180 degrees. The true angle is identified as the angle that produces the largest correlation coefficient across emissions. The magnitude of the velocity is found by cross-correlating consecutive lines focused along the direction of flow, to find the spatial shift corresponding to the velocity. In initial in vivo experiments, the method has however shown weaknesses by yielding outliers when a substantial clutter signal is present after clutter filtering. This is especially a problem when the flow angle is close to 90 degrees as the slow time frequencies from the flow signal is similar to that of the clutter, making clutter filtration difficult. When the angle determination fails, the correct velocity can no longer be found. The purpose of this work is to improve the robustness of the directional beamforming method, making precise in vivo measurement possible. A more robust angle estimator is proposed. Spatial averaging in the axial direction is applied over a depth of 2 wavelengths. Instead of traditionally beamforming a single line, three-lines are beamformed with an interline distance of a wavelength. To improve clutter filtering a post correlation clutter filter is applied, by removing peaks in the correlation functions corresponding to low velocities, since these peaks are believed not to be a result of the measured flow. The method has been tested on a flow phantom, using the RASMUS experimental scanner. The flow had a parabolic velocity distribution with a peak velocity of 0.1 m/s, and a flow angle of 90 degrees. The measurement were made with a 6.2 MHz linear array transducer, using 30 emissions and 128 transducer elements for each estimate. Using the same measurement setup, an initial in vivo study has been carried out. The measurements have been performed on the carotid artery of 11 human volunteers. To validate the method MR angiography has been performed on all human volunteers as a gold standard. For the phantom measurement 76.30 % of the angle estimates are within +- 5 degrees of the actual angle, when using the traditional setup. Using our new approach 98.32 % of the angle estimates are within +-5 degrees from the true angle. The comparison between the stroke volume measurements in the carotid artery calculated using directional beamforming and MR angiography, gives a correlation coeficient of 0.84. Phantom and in vivo measurements has been carried out with a more robust implementation of the directional beamforming method. With the applied changes, the method has shown improved results for in vivo measurements.