Flight delay prediction remains an important research topic due to its dynamic nature. Dynamic data-driven approach might provide a solution to this problem. To apply the approach, a flight delay state-space model is required to represent relationship among system states, as well as relationship between system states and input/output variables. Based on the analysis of delay event sequence, a state-space model was established and the input variable was studied. A genetic EM algorithm was applied to obtain global optimal estimates of parameters used in the mode. Validation based on probability interval tests shows that: the model has reasonable goodness of fit to the historical flight data, and the search performance of traditional EM algorithm can be improved by ideals of Genetic Algorithm.