Neural responses in sensory systems are typically triggered by a multitude
of stimulus features. Using information theory, we study the encoding accur
acy of a population of stochastically spiking neurons characterized by diff
erent tuning widths for the different features. The optimal encoding strate
gy for representing one feature most accurately consists of narrow tuning i
n the dimension to be encoded, to increase the single-neuron Fisher informa
tion, and broad tuning in all other dimensions, to increase the number of a
ctive neurons. Extremely narrow tuning without sufficient receptive field o
verlap will severely worsen the coding. This implies the existence of an op
timal tuning width for the feature to be encoded. Empirically, only a subse
t of all stimulus features will normally be accessible. In this case, relat
ive encoding errors can be calculated that yield a criterion for the functi
on of a neural population based on the measured tuning curves.