Abstract: A simple and general scheme is presented for establishing a threshold auto-regressive (TAR) model. Both threshold values and auto-regressive coefficients can be optimized with the genetic algorithm based on real coding genetic algorithm improved by the authors,and the difficulty problem of modeling of TAR is resolved,which gives a strong tool for widely using TAR model in meteorological forecasting. The calculation example shows that TAR model can effectively utilize the important information of the section interdependence during the meteorological time series observation dates by controlling threshold valves,reduce model errors,and ensure good stability and accuracy of the model forecasting. As a general method,the scheme has major theoretic valve and wide-ranging application to predicting of meteorological nonlinear time series.