We present a method and case study to predict and map the likelihood of wildfires spreading to the urban interface through statistical analysis of past fire patterns using 15 000 lines from 677 fires with known ignition points and date and random potential end points on the urban interface of Sydney, Australia. A binomial regression approach was used to model whether the fire burnt to the end point of the lines as a function of measures of distance, fuel, weather and barriers to spread. Fire weather had the strongest influence on burning likelihood followed by the percentage of the line that was forested, distance and time since last fire. Fuel treatments would substantially reduce risk from fires starting 1-4 km away from the interface. The model captured 90% of variation in burning with 98% predictive accuracy on test data and was not affected by spatial autocorrelation. We apply the method to map fire risk in Sydney and discuss how the method could be expanded to estimate total risk (from ignition to impact on assets). The method has considerable promise for predicting risk, especially as a complement to simulation methods.