Abstract: In order to construct a valid road weather warning model, the effect of weather on the daily number of all reported successive traffic accidents day by day from January 2002 to December 2004 (1096 valid samples) in the area of Xi'an city is examined. Thirteen meteorological elements sampled over the same area day by day are applied for the corresponding three years. The 1096 sample days are divided into two half-year parts given different seasonal regimes of the meteorological elements. Data are processed using statistical product and service solution (SPSS) version 12.0.Factor analysis methods are utilized to summarize four public factors, which are F_i(i=1, …, 4) and H_i(i=1, …, 4), integrated by temperature, visibility, relative humidity and rainfall in spring-summer half-year period and temperature, visibility, rainfall and pressure respectively in autumn-winter half-year period. Multi-collinearity is effectively overcome by this process and the meteorological variables are reduced from thirteen elements to four public factors.Logistic regression analysis based on these four public factors is applied. Initially the model is calculated, the number in the set that occurs most frequently in all traffic crashes data is obtained. Then it is supposed that if the raw crash data are greater than the mode value 1 is set to this variable and targets the high crashes frequency; if the raw data are smaller than the mode value 0 is applied to this variable and targets low crashes frequency. Binary logistic by entrance option is applied to get the road weather warning model over the two half-year periods. But F₁ (temperature factor in spring-summer period) and H₄(pressure factor in autumn-winter period) are shown to be insignificant, so they are abandoned when the logistic regression equations are constructed.The model is evaluated for preferable prediction accuracy through samples tests. The results show that the meteorological factors used to indicate their influences and forecast traffic crashes in Xi'an city result in a suitable and effective road weather warning model.In summary, according to significance level in the logistic regression equation and the value of factor loadings, the factors are respectively integrated by visibility, relative humidity and rainfall in spring-summer half-year period and temperature, visibility and rainfall in autumn-winter half-year period.