The fault coverages achieved by the test generation procedures based on genetic algorithms are smaller than the deterministic test generation procedures for combinational circuits.One of the possible causes for this deficiency is the high

long-distance schema

which exists in the process of automatic test pattern generation.Thus genetic algorithms in dealing with such problems easily fall into local optima or premature convergence.In this work

we firstly propose the test generation procedures based on estimation of distribution algorithms.Estimation of distribution algorithms are able to capture the interrelations between the primary inputs by joint probability distribution.And therefore obviate the influence of the high

long-distance schema;alleviate the problem of premature convergence.The experimental results for benchmark circuits prove that the proposed procedure can achieve higher fault coverage.