The split-spectrum amplitude-decorrelation angiography algorithm was optimized on a spectral optical coherence tomography system using a flow phantom. The number of times the spectrum was split and the bandwidth of each split were adjusted to maximize the flow phantom decorrelation signal-to-noise ratio. The improvement in flow detection was then demonstrated with en face retinal angiograms. The optimized algorithm increased the detectable retinal microvascular flow and decreased the variability of the quantified vessel density in OCT retinal angiograms of healthy human subjects.