Several aspects of the Color Rendering Index (CRI) are flawed, limiting its usefulness in assessing the color rendering capabilities of LEDs for general illumination. At NIST, we are developing recommendations to modify the CRI that would overcome these problems. The current CRI is based on only eight reflective samples, all of which are low to medium chromatic saturation. These colors do not adequately span the range of normal object colors. Some lights that are able to accurately render colors of low saturation perform poorly with highly saturated colors. This is particularly prominent with light sources with peaked spectral distributions as realized by solid-state lighting. We have assembled 15 Munsell samples that overcome these problems and have performed analysis to show the improvement. Additionally, the CRI penalizes lamps for showing increases in object chromatic saturation compared to reference lights, which is actually desirable for most applications. We suggest a new computation scheme for determining the color rendering score that differentiates between hue and saturation shifts and takes their directions into account. The uniform color space used in the CRI is outdated and a replacement will be recommended. The CRI matches the CCT of the reference to that of the test light. This can be problematic when lights are substantially bluish or reddish. Lights of extreme CCTs are frequently poor color renderers, though they can score very high on the current CRI. An improved chromatic adaptation correction calculation would eliminate the need to match CCT and an updated correction is being considered.