本文将函数逼近的方法引入到学习式搜索中，使学习式搜索通过一定数量的解题训练后可以建立起一个任意一致逼近理想函数ｈ＊（·）的启发估计函数ｈ（·）．本文给出了一个这样的学习式搜索算法Ａ－Ｂｎ，并证明了当训练例子集充分大后，Ａ－Ｂｎ可在多项式复杂度内解决任一后来提交的同类问题． In this paper, the method of function approximation is introduced into the learning search, so that the learning search can establish a heuristic estimation function h (·) of an arbitrarily uniformly approximated ideal function h * (·) after solving a certain number of problems. In this paper, we present such a learning search algorithm A-Bn, and prove that A-Bn can solve any similar problem submitted later in polynomial complexity when the training example set is sufficiently large.