This talk explains and applies a diagnostic framework designed to understand the origin of tropical cyclone (TC) track errors. I will focus on recent cases (Sandy, Joaquin, Lionrock and Matthew) where particularly large errors occurred in forecasts from the NCEP Global Forecast System (GFS). The diagnostic partitions errors in the steering flow, defined as the horizontally and vertically averaged wind that best matches the motion of the TC, into environmental wind errors and errors in the horizontal or vertical structure of the TC. Particularly large forecast track errors result when steering-flow errors occur in conjunction with strongly diffluent environmental wind fields. Results also point to a systematic error in the GFS that may explain why GFS forecasts occasionally produce much larger errors than forecasts from the ECMWF model.