Since they are seldom used as analog elements in previous analog forecast method, the numerical model integral products are used in the third level forecast in SMAT(i.e. synthetica- lly multilevel analog forecast technology) in this paper. This method includes\" the reducing FAR (mean vacant-forecast rate) idea\", which is possessed of many advantages in filtering analog term. In SMAT, similitude degree between samples is described by meteorological elements combina- tion, which makes further progress than using single element to describe similitude in earlier analog forecast. In the paper, model output data, air sounding data, surface observation data and weather map data from 1990 to 2002 are applied to perform SMAT experiment of quasi- stationary front rainstorm. Important trial results are provided as follows: Forecast successful index (CSI,0.36) is improved one point. Both forecast precise rate(COR,0.59) and lost-forecast rate (POD,0.33) are also better than other forecast methods. Results show that, based on model output data, SMAT can perform more accurately forecast of quasi-stationary front rainstorm. The optimal analog elements reveal triggering mechanisms locate in lower troposphere while upper level systems are more important in maintaining phase, and these analog variables should be firstly took account in operational forecast of quasi-stationary front rainstorm. In addition, results reveal the position of key zone is mainly decided by the position of system caused rainstorm.