This paper uses estimation techniques related to those of Galbraith and Zinde-Walsh (2000) for ARCH and GARCH models, based on realized volatility (Andersen and Bollerslev 1998, and others), to estimate the conditional quantiles of daily volatility in samples of equity index and foreign exchange data. These techniques in principle allow us to characterize the entire conditional distribution of volatility, conditioning on past realized volatility and past squared returns. We take samples of daily and intra-day returns on the Toronto Stock Exchange 35 index, the DM/$ US exchange rate and the Yen/$ US exchange rate. In addition to information about the conditional extremes of volatility, we find some evidence that lower percentiles of the conditional distribution rise proportionately less in high-volatility periods than do the higher percentiles.