Fisher information is used to analyze the accuracy with which a neural popu
lation encodes D stimulus features. It turns out that the form of response
variability has a major impact on the encoding capacity and therefore plays
an important role in the selection of an appropriate neural model. In part
icular, in the presence of baseline firing, the reconstruction error rapidl
y increases with D in the case of Poissonian noise but not for additive noi
se. The existence of limited-range correlations of the type found in cortic
al tissue yields a saturation of the Fisher information content as a functi
on of the population size only for an additive noise model. We also show th
at random variability in the correlation coefficient within a neural popula
tion, as found empirically, considerably improves the average encoding qual
ity. Finally, the representational accuracy of populations with inhomogeneo
us tuning properties, either with variability in the tuning widths or fragm
ented into specialized subpopulations, is superior to the case of identical
and radially symmetric tuning curves usually considered in the literature.