A major difficulty in hurricane forecasting is the specification of accurate initial conditions that are dynamically consistent with the numerical model. Furthermore, a large quantity of available data are not fully exploited in current operational data assimilation schemes. As an initial step towards improving data assimilation in tropical cyclone forecasts, a 4d-Var scheme is used to assimilate QuikScat and AMSU satellite data into the MM5 model in the vicinity of the incipient Hurricane Floyd. Comparisons of the 4d-Var results versus existing high- and low- resolution MM5 simulations and 3d-Var assimilations will be presented.