The performance of conventional CFM imaging will often be degraded due to the relatively low number of pulses (4-10) used for each velocity estimate. To circumvent this problem we propose a new method using Frequency Splitting (FS). The FS method uses broad band chirps as excitation pulses instead of narrow band pulses as in conventional CFM imaging. By appropriate filtration, the returned signals are divided into a number of narrow band signals which are approximately disjoint. After clutter filtering the velocities are found from each frequency band using a conventional autocorrelation estimator. Finally the velocity estimates from each frequency band are averaged to obtain an improved velocity estimate. The FS method has been evaluated in simulations using the Field II program and in flow phantom experiments using the experimental ultrasound scanner RASMUS. In both simulations and experiments, a 5 MHz linear array transducer was used to scan a vessel situated at 30 mm depth with a maximum flow velocity of 0.1 m/s. The pulse repetition frequency was 1.8 kHz and the angle between the flow and the beam was 60 deg. A 15 μs chirp was used as excitation pulse and 40 independent velocity estimates were obtained using the FS method with 10 pulse transmissions used for each estimate. For comparison, a 8 cycles sinusoid pulse at 5 MHz was used to acquire 40 independent velocity estimates, each derived from 10 pulse emissions. Here the velocity was found using a conventional autocorrelation estimator. In the simulation, the relative mean standard deviation of the velocity estimates over the vessel was 2.43% when using the FD method and the relative mean absolute bias was 1.84%. For the reference 8 oscillation pulse, the relative mean standard deviation over the vessel was 4.91% and the relative mean absolute bias was 1.78%. In the experiments the relative mean standard deviation of the velocity estimates over the vessel was 2.41% when using the FD method and the relative mean absolute bias was 1.56%. For the reference 8 oscillation pulse, the relative mean standard deviation over the vessel was 4.76% and the relative mean absolute bias was 3.12%.