The main defect is that it assumes normality of log returns (geometric brownian motion), whereas there is clear empirical evidence that returns actually have fatter tailed distribution (higher kurtosis: extreme events are more likely to occur than in B-S model)