We suggest that an index of the temporal variability of abundances should measure proportional change. SD (log[X]) seems to provide a general and interpretable statistic. However, when based on raw counts, a variability measure incorporates sampling error as well as the variability of the underlying densities. If counts have been replicated at each time period, then separating these may be a relatively straightforward process. If counts have not been replicated the problem is more difficult and can only be solved under particular assumptions. We outline two such solutions. Considering variabilities in the light of a model which discriminates sampling error assists in the interpretation of some measures of spatial variability, and resolves much of the recent controversy over the fitting of variance-mean relationships.