Microstructure modeling of aluminum alloy castings has been carried out by using a new stochastic modeling methods coupled with macro heat transfer calculations. The actual dendritic shape is substituted for the square envelope in the cellular automaton model. A physical model and a mathematical model for the simplified dendrite shape are established in the paper, in which a shape function is presented to describe the dendrite shape contour. On the basis of the simplified grain shape, a coordinate transformation method is used to describe the equiaxed grain growth in the undercooled melts and the node capturing during the further growth of a grain. A continuous nucleation model is applied to deal with heterogeneous nucleation phenomenona. Dendrite tip growth kinetics and preferential <100> crystallographic orientation are taken into account. The stochastic nature of nucleation process as well as the deterministic of dendrite growth is considered to simulate the crystal growth. A microstructure simulation scheme is developed to model the grain formation according to the above model. Two- dimensional calculations are performed to simulate the evolution of equiaxed dendritic morphologies. In order to verify the modeling results, sample castings are cast in sand molds and metal molds. Simulation results show that grain size is smaller for metal mold casting, but larger for sand mold casting, in good agreement with both the experimental results and the postulated solidification mechanism.