Web applications often suffer from interaction problems, causing users to struggle to fill out forms or find information. To diagnose these problems without incurring large costs, a popular solution is to analyze interaction events automatically to find problematic behavior patterns. However, these analyses often do not take into account the user’s interaction style, such as speed. In this paper we present a study on the web users’ speed by using recordings of real sessions to find how early this characteristic, measured in terms of mouse, scroll and keyboard speed, can be estimated. Preliminary results suggest that it takes approximately 50 seconds or 500 events to obtain a close estimate (>0.85) of the benchmark speed, calculated from the total number of events in the session.