Aim: This paper describes the characteristics of the spatio-temporal distribution of vegetation fires as detected from satellite data for the 12 months April 1992 to March 1993. Location: Fires are detected daily at a spatial resolution of 1 km for all land areas of the globe. Methods: From the fire location information a daily gridded product at 0.5 degrees by 0.5 degrees has been constructed. Two methods of characterizing the spatio-temporal pattern of vegetation fires are discussed. The first applies empirical orthogonal function analysis to the monthly series of gridded data. The second approach defines and extracts a number of spatial and temporal parameters from the gridded product. The descriptive parameters extracted are used in a cluster analysis in order to group cells with similar characteristics into a small number of classes. Results: Using daily global satellite observations, it is possible to characterize the spatial and temporal variability in fire activity. Most of this variability is within the tropical belt, where the majority of fire activity is concentrated, nonetheless fire was also detected in temperate and boreal regions. The period in which fire occurred varied from region to region. Parameterization provided a very synthetic view of this variability facilitating regional intercomparison. Clustering identifies five classes of fire activity, each of which can be associated with particular climatic conditions, vegetation types and land-use. Main conclusions: Global monitoring of vegetation fire from satellite is possible. The analysis provides a coherent, consistent and synoptic view of global fire activity with one data set. The type of information extracted can be of use in global atmospheric chemistry modeling and for studying the role of fire in relation to global change issues.