Aim: Fire is a globally important disturbance that affects nearly all vegetated biomes. Previous regional studies have suggested that the predictable seasonal pattern of a climatic time series, or seasonality, might aid in the prediction of average fire activity, but it is not known whether these findings are applicable globally. Here, we investigate how seasonality can be used to explain variations in fire activity on a global scale. Location: Global, 60° S–60° N. Time period: Data averaged over the period 1999-2015. Methods: We describe a method to partition a periodic seasonal cycle into two seasons and define conceptually simple temporal metrics that describe spatial variability in seasonality. We explore the usefulness of these metrics in explaining global fire activity using the average monthly time series of precipitation and temperature and a flexible machine learning procedure (random forests). Results: A simple model that uses only precipitation and temperature amplitude and synchrony between wet and warm seasons correctly predicts 66% of the variability in global fire activity, substantially more than a model with mean annual temperature and precipitation. A more complex model that includes all nine metrics predicts 87% of variability in global fire activity. Main conclusions: This study shows that seasonality of temperature and precipitation can be used to predict multi‐year average fire activity in a globally relevant way. The mechanisms highlighted in our work could be used to improve global fire models and enhance their ability to represent the spatial patterns of fire activity. Our method might also be useful in hindcasting historical fire using reanalysis or predicting future fire regimes using coarse output from climate models.