A new class of fractionally integrated GARCH and EGARCH models for character-izing financial market volatility is discussed. Monte Carlo simulations illustrate the reliability of quasi maximum likelihood estimation methods, standard model selection criteria, and residual-based portmanteau diagnostic tests in this context. New empirical evidence suggests that the apparent long-run dependence in U.S. stock market volatility is best described by a mean-reverting fractionally integrated process, so that a shock to the optimal forecast of the future conditional variance dissipate at a slow hyperbolic rate. The asset pricing implications of this finding is illustrated via the implementation of various option pricing formula.