FST, a measurement of the genetic differentiation among subpopulations, is a fundamental parameter in population genetics, with many valuable applications in molecular biology, evolutionary biology, conservation and forensics. One of its close relatives, GST, has been widely used to measure differentiation from highly polymorphic markers such as microsatellites. However, because of the high mutation rate of such markers, GST may underestimate the genomic differentiation due to demographic causes such as migration rate and subpopulation size. A new statistic proposed recently, Jost's D, was claimed to have better properties than GST and was advocated to replace GST as a measure of differentiation. This paper shows that D is not a proper measure of differentiation because it fails to meet some fundamental requirements as a differentiation statistic, and is hardly estimable without bias in practice. D is highly dependent on the gene diversity of a marker and on the unknown parameter of the number of subpopulations, is highly sensitive to how alleles and loci are defined and how data are analysed, does not increase monotonically with either divergence time or drift, and does not always have a maximal value of 1. The maximal D value can be zero or close to zero, depending on the number of alleles at a locus relative to the number of subpopulations. I suggest continuing the use of GST, with caution in its interpretation when highly polymorphic markers are used, before a better estimator of FST that explicitly accounts for mutations is developed.