Given a grammar compressed string S, a pattern P, and d >= 0, we consider the problem of finding all occurrences of P' in S such that d(P, P') <= d with respect to Hamming distance. We propose an algorithm for this problem in 0(1g lg n 1g* N(m+ d occdlg 1g N)) time, where N = vertical bar S vertical bar, m = Ik n is the number of variables in the grammar compression, and occ(d) is the frequency of an evidence of a substring of P. We implement this algorithm and compare with a naive filtering on the grammar compression.