Neural networks are frequently used as adaptive classifiers. This rese
arch represents an attempt to measure the ''neural complexity'' of any
regular set of binary strings, that is, to quantify the size of a rec
urrent continuous-valued neural network that is needed for correctly c
lassifying the given regular set. Our estimate provides a predictor th
at is superior to the size of the minimal automaton that was used as a
n upper bound so far. Moreover, it is easily computable, using techniq
ues from the theory of rational power series in non-commuting variable
s.