A database, the largest one to date, of magnetosheath encounters by geosynchronous satellites during 1986-1992 and 1999-2000 and upstream observations by Wind, IMP 8, or Geotail in the solar wind are used to estimate the forecasting capability of the models of Chao et al. , Shue et al. , and Petrinec and Russell . For each of the 1-min resolution data points obtained by GOES spacecraft, we check the following two things: if the magnetosheath was observed by the spacecraft, and if each of the three models predicted a magnetosheath encounter by the spacecraft. Three parameters are defined to quantify the models' forecasting capability: probability of prediction (PoP), probability of detection (PoD), and false alarm rate (FAR). A higher PoP and PoD with a lower FAR imply a better forecasting model. In the 1986-1992 period we found that most of the magnetosheath encounters observed at 6.6 RE are detected. In particular, the Chao et al.  model predicts the lowest FAR compared with those of the other two models. We have also studied the magnetosheath encounters made by GOES spacecraft for the period from 1999 to 2000 using Wind and Geotail as the solar wind monitor. This independent database of magnetosheath encounters during 1999-2000 confirms our previous anticipations. The PoD of Petrinec and Russell  model (94%) is much higher than those of Shue et al.  (74%) and Chao et al.  (84%) models. The Chao et al.  model has a higher PoP than the other two models. The values of FAR for Chao et al. , Shue et al. , and Petrinec and Russell  models are 25, 27, and 40%, respectively.