A new variance estimate for the Mantel-Haenszel log-odds-ratio is proposed and compared to the proposals of Hauck, Breslow, and Flanders in two special settings where reasonable variance estimates are known. The new variance estimate is symmetric in the sense proposed by Ury and is derived in a similar manner to the proposal of Flanders. A second variance estimate that is only valid for the special case of (1,M)-matching is also discussed and compared to the more general estimator. (14pp.)