This study investigates the emergence of creoles through computer simulations of language contact in an artificial community. Creoles are full-fledged new languages which infants growing in a multilingual community acquire as their native language. We define a creole in the agent community as a new language dominating the community. We have developed an agent-based model with Kirby (2002)'s Iterative Learning Model where agents are connected with neighbors on a social network. The infant agents acquire a compositional grammar, hearing utterances from their parents and neighbors. Assuming language exposure plays a key role for creolization, we introduce a parameter of exposure to neighbors to the model. Language groups result from clustering similar languages, each of which is spoken by an agent. Experimental results show a certain degree of the exposure is necessary for creolization. Further investigation suggests that the structure of the social network affects that of agents' grammars.