The statistical properties of spike trains generated by a sensory neur
on are studied. It is shown that the spike trains exhibit universal st
atistical behavior over short times, modulated by a strongly stimulus-
dependent behavior over long times. This decomposition is accounted fo
r by a ''frequency integrator'' model, under conditions of time scale
separation. We provide explicit formulas for the statistical propertie
s in both the universal and the stimulus-dependent regimes, which are
in very good agreement with the data, The universal regime is characte
rized by a dimensionless free parameter, which is observed experimenta
lly to remain constant under different external stimuli. [S0031-9007(9
8)07566-8].