A trust measurement method called certified belief in strength (CBS) for Extreme Learning Machine (ELM) Multi Agent Systems (MAS) is proposed in this paper. The CBS method is used to improve the performance of the individual agents of the MAS, i.e., ELM neural network. Then, trust measurement is achieved based on reputation and strength of the individual agents. In addition, trust is assemble from strong elements that are associated with the CBS which let the ELM to improve the performance of the MAS. The efficiency of the ELM-MAS-CBS model is verified with several activation function using benchmark datasets which are Pima Indians Diabetes (PID), Iris and Wine. The results show that the proposed ELM-MAS-CBS model is able to achieve better accuracy as compared with other approaches.