This paper presents a numerical model for the prediction of the minimum ignition temperature (MIT) of dust clouds. First, a physical model is developed for the dust cloud ignition in the Godbert-Greenwald furnace. A numerical approach is then applied for the MIT prediction based on the physical model. The model considers heat transfer between the air and dust particles, the dust particle reaction kinetics, and the residence times of dust clouds in the furnace. In general, for the 13 dusts studied, the calculated MIT data are in agreement with the experimental values. There is also great accordance between the experimental and numerical MIT variation trends against particle size. Two different ignition modes are discovered. The first one consists in ignition near the furnace wall for bigger particles characterized by rather short residence times. In the second mode, the ignition starts from the center of the furnace by self-heating of the dust cloud for smaller particles with longer residence times. For magnesium, as dust concentration increases, the lowest ignition temperature of the dust cloud IT(conc) decreases first, then transits to increase at a certain point. The transition happens at different dust concentrations for different particle sizes. Moreover, the MIT of the magnesium dust cloud generally increases as particle size increases, but the increasing trend stagnates within a certain medium particle size range.