A method of increasing the sampling efficiency in Monte Carlo calculations of thick shield penetration has been developed. The procedure alters the effective mean-free-path in such a way as to maximize the rate of convergence of the transmission probability. The approach is semiempirical in nature and has been shown to be remarkably insensitive to geometry. The primary dependence appears to be on the nonabsorption probability at each collision, with secondary dependence on the distance to escape. The procedure is simple enough to permit its incorporation into existing Monte Carlo codes with a minimum of programming effort.