Reconnaissance data including dropsonde data and airborne Doppler radar data have been used for data assimilation in GFDL hurricane forecast model. Preliminary results show that including reconnaissance data in the data analysis gives more realistic initial conditions in the GFDL hurricane vortex compared to the bogus scheme. We will present case studies and show the effects of the new analysis fields on the hurricane track and intensity forecasts in the GFDL model. This effort is a new initiative at NCEP to initialize the hurricane vortex with core observations for next generation high resolution hurricane prediction.