Many practitioners point out that the speculative profits of institutional traders arc eroded by the difficulty in gauging the price impact of their trades. In this paper. we develop a model of strategic trading where speculators face such a dilemma because of incomplete information about time-varying market liquidity. Unlike the competitive market makers that they trade against, informed traders do not know whether the liquidity ( "noise") trades are generated from a distribution with high or low variance. Instead, they have to learn about liquidity from past prices and trading volume. Extreme price deviations from forecasts of fundamentaIs based on public news or low trading volume tend to lead to revisions of beliefs in favor of the low liquidity state. This revision in beliefs implies that strategie trades and market statistics such as informational efficiency arc path-dependent on past market outcomes. Our paper has a number of normative implications for practitioners concerned with gauging the potential price impact of their trades.