Abstract: The emergence of the small probability of severe weather events is attributed to specific factor combinations of some early meteorological elements. The nonlinear and complicated characteristics of factor combinations can be used to find the relationship between forecast object and forecast factors. Based on this method, the relationship between hail events in the south of Sichuan Basin and some meteorological elements calculated by the sounding data is investigated, and a hail forecast index discriminant is established. The discriminant is physically significant and applied in daily operation.Hailstorm is a meso-scale weather system with the temporal scale of several to dozens of hours, and the horizontal scale of several hundred kilometers. In real business, the T-lnp sounding data are observed at 0800 BT and 2000 BT every day, and the hail forecasting is carried out every 12 hours. A sample sets of 7 hail events and 38 non-hail events near Yibin Station is established. Using the T-lnp sounding data, 3422 meteorological elements are calculated as forecast factors, including temperature, height, moisture, saturation vapor pressure, potential pseudo-equivalent temperature, K index and so on. Based on factors combination analysis method, 2 main factors and 2 conditional factors are selected from 3422 meteorological elements and their critical values are calculated. The main factors are T_σ400^*-T_σ850 and G_z400-G_zsurface, and the conditional factors are e₇₀₀-e_s700 and T_d700-T_σ700^*, T_σ400^* stands for saturated wet static temperature at 400 hPa, and T_σ850 stands for wet static temperature at 850 hPa; G_z400 and G_zsurface stand for vertical pressure gradient at 400 hPa and the surface level; e₇₀₀ and e_s700 stand for vapour pressure and saturated vapour pressure at 700 hPa; T_d700 and T_σ700^* stand for dew point temperature and saturated wet static temperature at 700 hPa. The hail forecast indexes discriminant nearby Yibin Station is established using these data.The environmental state of hailstorm generation and the unstable mechanism of severe convective weather can be explained by the hail forecast indexes discriminant, which is evaluated using historical records of the year of 2008. Among 65 warnings, the real hail events never miss but the false alarm ratio reaches 67.7%, which should be further distinguished using radar observations. The overall probability of detection is 84%, and the critical success index is 30.4%. The result shows that factor combination analysis method is feasible to some extent.