This paper examines a persistent problem with the NCEP Medium Range Forecast (MRF) model in the tropics. During the mid1990s, the model produced a series of spurious smallscale spinups in the tropics. Despite corrections, the model now develops very realistic looking, yet still fictitious, tropical cyclones (TCs). A series of model runs from June 1998 highlights the problem. Not only did the model develop a large number of imaginary TCs in the Eastern Pacific, but four of them actually appeared in the model initial analysis. One part of the problem may arise from a tooefficient convective parameterization, as these spurious TCs form mainly during episodes of enhanced convection. Another part is possible shortcomings in the model analysis, with insufficient low level data reaching the model to contradict its first guess field. This problem has serious implications for TC track forecasting since the Aviation run of the MRF serves as a background field for most of the TPC/NHC track prediction models