Grassland fire is a major disturbance to ecosystems and economies throughout the world. Research on the spatial pattern of grassland fire is therefore important in understanding the dynamics of fire disturbance and providing evidence for fire management and prevention. In this study we used a spatial point process modeling approach to study the factors contributing to fire occurrence in the Hulunbeir grassland of the Inner Mongolia Autonomous Region. In previous studies, Ripley' K function, Kernel density and Poisson model have been used in the studies of spatial-temporal pattern of forest fires. But the distribution pattern of grassland fires was usually described by overlaying fire points on top of the administrative districts or study regions. The properties of spatial distribution, such as clustering, dispersion, randomness, were often omitted. In this study, Repley's K function was used to investigate the spatial distribution pattern of human-caused fires in the Hulunbeir grassland. The distribution of fire locations was found to be spatially clustered in the months of fire season and between years. The distances of spatial cluster distribution were less than 250km, 265km, 245km, 200km and 245km in April, May, June, September and October respectively. The statistical test showed that the cluster distributions were significant except for October. The distances of spatial cluster distribution were less than 210km, 280km, 260km, 220km and 220km in 1976-1980, 1981-1984, 1985-1988, 1989-192 and 1993-1996 respectively. All the cluster distributions were significant. The spatial distribution densities of human-caused grassland fires and the impacting factors including residential locations, roads and fields were calculated by the function of Kernel density. The results showed that human-caused grassland fires were widely distributed and the main "hot-spot" was in the extent of 123.05-124.82°E, 48.25-50.21°N. The second "hot-spot" had four regions and the center locations were: 117.52°E,49.51°N; 120.73°E,49.24°N;121.94°E,48.74°N;123.73°E,50.58°N. The "hot-spots" of residential points and fields were distributed in the east and middle of the study area. The "hot-spot" of roads was more widespread than the residential points and fields. The spatial distribution density of human-caused grassland fires was significantly correlated with the impacting factors (residential points, roads, fields, weather and topography) and the Pearson coefficients were 0.448, 0.236, 0.602, 0.161 and -0.042 respectively (P<0.001). The spatial patterns of fire sources (residential points, roads and fields) are found to be more closely associated with the spatial density of human-caused grassland fires. The correlation of weather with the spatial pattern of grassland fire was at a lower degree than these of residential points, roads and fields. It showed that the weather condition was not a limiting factor in affecting the occurrence of human-caused grassland fire in the fire season. Overall, the most important factors impacted human-caused grassland fire occurrence were the spatial distribution of fire sources (residential points, roads and fields). Topography has negative correlation with grassland fire occurrence. The reason may be that the human activities decreased with the increase of slope. In each month of fire seasons, the spatial distribution pattern of human-caused grassland fires was also mainly impacted by the distribution of fire source (residential points, roads and fields). The methods developed in this study can be applied to predict human-caused grassland fire risk, aiding decisions in preventative management strategies of grassland fires.