Collective behavior refers to social processes and events which do not reflect existing social structure (laws, conventions, and institutions), but which emerge in a”spontaneous” way. It is a common phenomenon in microbiology: a group of cells can spontaneously form different structures under different conditions. How cells interact with each other and achieve this kind of coordinated cell movement is of active scientific interest. As a model organism for bacterial collective behavior,Myxococcus xanthus is widely studied to uncover the mechanism behind bacterial collective behavior. In this work, we applied agent-based models to study the aggregation behavior of M. xanthus cells under starvation and the important cell behaviors for csgA and pilC mutants aggregation.Experiments have shown that WT M. xanthus cells perform a biased walk to-wards aggregation center and this biased walk helps aggregation. To uncover the mechanism of the biased walk, we first developed a model where each cell is modeled as an agent, represented by a point-particle and characterized by its position and moving direction. At low agent density, the model recapitulates the dynamic pat-terns observed by experiments and a previous biophysical model. At high cell density,we extended the model based on the experimental data of the biased movement to-wards aggregates. We tested two possible mechanisms for this biased movement and demonstrate that a chemotax is model with adaptation can reproduce the observed experimental results leading to the formation of stable aggregates. Furthermore, our model reproduces the experimentally observed patterns of cell alignment around aggregates. Next, we applied a data-driven agent-based model to investigate what cell behaviors are important for the rescue of aggregation in two mutants: csgA and pilC, which cannot aggregate unless mixed with wild type (WT) cells. We discovered that when mixed with WT cells, both mutants show biased movements and reduced motility inside aggregates. These behaviors are shown to be important to aggregation in our agent-based simulations. However, some mutant behaviors remain different from WT cells demonstrating that perfect recreation of WT behavior is unnecessary.This work proposes a possible mechanism of the aggregation of M. xanthus bacteria and has shown that some cell behaviors are more important than others in aggregation. Our agent-based model provides a general framework that can be used to study self-organization behaviors in other n other surface motile bacteria