Pricing American options is an interesting research topic since there is no ana- lytical solution to value these derivatives. Di¤erent numerical methods have been proposed in the literature with some, if not all, either limited to a speci c payo¤ or not applicable to multidimensional cases. Applications of Monte Carlo meth- ods to price American options is a relatively new area that started with Longsta¤ and Schwartz (2001). Since then, few variations of that methodology have been pro- posed. The general conclusion is that Monte Carlo estimators tend to underestimate the true option price. The present paper follows Glasserman and Yu (2004b) and proposes a novel Monte Carlo approach, based on designing "optimal martingales" to determine stopping times. We show that our martingale approach can also be used to compute the dual as described in Rogers (2002).