Can't recommend the Mandelbrot book highly enough - he was a student of Paul Levy, and he wrote extensively about why Gaussian is not a good choice to model financial time series.