AbstractA fire weather index (FWI) is developed using wildfire occurrence data and Automated Surface Observing System weather observations within a subregion of the northeastern United States (NEUS) from 1999 to 2008. Average values of several meteorological variables, including near-surface temperature, relative humidity, dewpoint, wind speed, and cumulative daily precipitation, are compared on observed wildfire days with their climatological average (“climatology”) using a bootstrap resampling approach. Average daily minimum relative humidity is significantly lower than climatology on wildfire occurrence days, and average daily maximum temperature and average daily maximum wind speed are slightly higher on wildfire occurrence days. Using the potentially important weather variables (relative humidity, temperature, and wind speed) as inputs, different formulations of a binomial logistic regression model are tested to assess the potential of these atmospheric variables for diagnosing the probability of wildfire occurrence. The FWI is defined using probabilistic output from the preferred binomial logistic regression configuration. Relative humidity and temperature are the only significant predictors in the binomial logistic regression. The binomial logistic regression model is reliable and has more probabilistic skill than climatology using an independent verification dataset. Using the binomial logistic regression output probabilities, an FWI is developed ranging from 0 (minimum potential) to 3 (high potential) and is verified independently for two separate subdomains within the NEUS. The climatology of the FWI reproduces observed fire occurrence probabilities between 1999 and 2008 over a subdomain of the NEUS.