If all of the data is positive, but the mean - 1*sigma is negative, the gaussian approximation probably isn't very good....