Through the example of partial barrier options, we show that accuracy in the tail of the bivariate normal distribution is critical. We then propose a small change to a popular algorithm for the bivariate normal distribution in order to increase its accuracy