We analyze the relationship between spreads and an indicator for information based transactions on trade-by-trade data. Classifying trades on the NYSE in six categories with respect to their volume relative to the quoted depth, we employ an ordered probit model to predict the category of a trade given the current market conditions. This approach allows us to test certain market microstructure hypothesis on the determinants of the buy-sell pressure. The difference between the predicted and the actual trade category (the surprise) is found to have explanatory power for the observed spreads beyond raw volume, volume relative to the quoted depth, and previous trading volume. The positive effect of the previous surprise on the observed spreads confirms the hypothesis that market-makers react to the increased probability of having traded with an informed trader by widening the spread.