In this paper, a predictive multi-scale model based on a cellular automaton (CA)-finite element (FE) method has been developed to simulate thermal history and microstructure evolution during metal solidification for the Direct Metal Deposition (DMD) process. The macroscopic FE calculation that is validated by thermocouple experiment is developed to simulate the transient temperature field and cooling rate of single layer and multiple layers. In order to integrate the different scales, a CA–FE coupled model is developed to combine with thermal history and simulate grain growth. In the mesoscopic CA model, heterogeneous nucleation sites, grain growth orientation and rate, epitaxial growth, re-melting of pre-existing grains, metal addition, grain competitive growth, and columnar to equiaxed phenomena are simulated. The CA model is able to show the entrapment of neighboring cells and the relationship between undercooling and the grain growth rate. The model predicts the grain size, and the morphological evolution during the solidification phase of the deposition process. The developed “decentered polygon” growth algorithm is appropriate for the non-uniform temperature field. Finally, the single and multiple-layer DMD experiment is conducted to validate the characteristics of grain features in the simulation.