We propose a characteristic representation of one-dimensional and two-
state, three-neighbor cellular automaton rules, which can be interpret
ed as a numerical description of Wolfram's classification scheme. Each
rule can be considered to behave, in many situations, like some other
rules. The similarity of behavior is quantified. The quantification i
s done by creating a function designed to give an average weight of co
nfigurations after many time steps, and then comparing this function,
in a precise way, to some other rules by the value of coefficients exp
anded by the rule functions. We call such an expansion form of the fun
ction characteristic representation. The simulation result of the repr
esentation shows that Class IV rules are distinguished from Class III
rules by the similarity to Class Il rules. Criticality of Class IV rul
es is displayed by functions linearly combining typical Class II and I
II rule functions.