Various degrees of residential segregation by income and race generally exist in U.S. cities. This study extends Sethi and Somanathan’s theoretical model (J Polit Econ 112:1296–1321, 2004) by presenting an agent-based sorting, repeated-game model to quantify the patterns of segregation from a broader perspective. Based on the belief that residential racial segregation is a probabilistic problem without assured results, a numerical model—calibrated to U.S. household income data—is proposed to examine residential segregation by income and racial preferences. Similar to the SimSeg model developed by Fosset (J Math Sociol 30:185–274, 2006a; J Math Sociol 35:114–145, 2011), the numerical model we construct is based on a simple format which also explores segregation dynamics. The simulation results exhibit various degrees of segregation probability in a hypothetical three-neighborhood scenario. It also reveals that although income plays an important role, racial consciousness—the measurement of an agent’s attitude toward the racial composition of the neighborhood—is the dominant factor in determining residential segregation.