In this paper we provide evidence for Evans and Lyons' (2005b) model of an information aggregation process in FX markets using a German bank's end-user order flow from 2002 to 2003. Though customer order flow is unambiguously the vehicle incorporating non-public information into exchange rates over time, our empirical analysis does not support the widespread optimism in the market microstructure literature that customer order flow is the high-powered source of information easily exploitable for short-run speculation. Moreover, commercial customers' order flow produces negative coefficients in contemporaneous return regressions, stressing their role as liquidity providers.