For most LPC-based low bit rate speech coders, the LPC coefficients must be transmitted. To minimize the distortion in encoding the LPC coefficients, both reflection coefficients and line spectral pair (LSP) coefficients have been used. It has been claimed that, for the same degree of distortion, encoding the LSP coefficients requires about 25% less bits than the encoding of the reflection coefficients. However, these comparisons were not carried out when both these coefficient sets were optimally quantized. It is intended that this paper provides such a comparison, with the quantizers and bit allocations both optimally derived using the same performance criterion. Our comparison shows that optimal scalar quantization of reflection and LSP coefficients produces approximately the same distortion. This paper also introduces a differential LSP quantization scheme which out-performs the more conventional LPC quantization schemes.