Chapada das Mesas National Park extends over an area of 160,046 ha in the municipalities of Carolina, Riachão, Estreito and Imperatriz in the south central region of the state of Maranhão, northeast Brazil, in a savanna-like biome known as the Cerrado. The park has a rich biodiversity, making the need for conservation all the more important. The weather conditions in the region increase the likelihood of wildfires, so that a monitoring and control system for the area is needed to help conservation efforts. This article proposes a methodology that uses data-mining techniques to predict outbreaks of wildfires in the park some hours in advance. Predictive models using wildfire records and a meteorology dataset for 11 months in 2010 were built. Two different classification techniques for predicting wildfires were used: artificial neural networks and classification rules. The two models built with these techniques showed good accuracy when tested with the validation samples and could be used as additional tools for predicting the risk of fires in the area.