This paper presents an empirical model for inferring the private information content of trades at the transaction level. The trade-indicator model of Glosten and Harris (1988) is extended to a two-state regime-switching setting, and the model is estimated using tick-by-tick data from the New York Stock Exchange (NYSE). The specialist is found to react in accordance with the proposed model. Bid-ask quotes set after the execution of a trade reflect the conjectured information content of that particular trade. Based on the estimated model four empirical results emerge: (a) the suggested regime-switching model fit data well; (b) the reverse J-shaped pattern of intra-daily quoted spreads is shown to agree with the clustering of costs incurred by the specialist through trading with better-informed agents; (c) on average 9% of all trades are found to reveal private: information to the specialist; (d) results regarding the trading volume of informed traders support the stealth trading hypothesis suggested by Barclay and Warner (1993).