During thunderstorm periods, terminal air traffic planners make a number of key decisions. They decide when to close and re-open arrival fixes, departure fixes, and runways; they anticipate and execute changes in runway configuration; they negotiate routing and flow rate decisions with Air Route Traffic Control Center (ART CC) traffic managers; and they set the airport acceptance rate. In making each of these decisions, the traffic planner looks at a weather radar display and makes an educated guess at answering the two following questions: - What will the weather be like in the airspace and time period in question? - Will the pilots be able and willing to fly through that airspace during that time? The same two questions will be important for advanced terminal automation systems. One key element of air traffic automation systems such as the Center-TRACON Automation System (CTAS) is the calculation of candidate trajectories for each aircraft for the time period of automation control. To make this calculation, the automation software must know which routes will be usable during the control period. The first of the two fundamental questions is being addressed by the convective weather Product Development Team (PDT) of the FAA's Aviation Weather Research program. (Wolfson, 1997; Wolfson, 1999; Hallowell, 1999; Forman, 1999; Evans, 1997) The second fundamental question is the subject of the work reported here. The state of the art answer to the second question is a widely quoted air traffic control rule-of-thumb which says that pilots generally do not penetrate precipitation that is NWS VIP level 3 (i.e. 41 dBZ) or higher. That is not to say that air traffic controllers always vector aircraft around level 3+ cells but rather that they begin to anticipate pilot requests for deviations when the weather approaches level 3. A suite of new weather sensors have become available that provide much more comprehensive information on convective weather features than was available in the past. Additionally, flight-related data such as preceding pilot behavior and whether a flight is running late are easier to obtain than in the past. In this study we develop an objective quantitative assessment of which weather and flight-related variables best explain pilot deviation decision-making.