A variety of web services have emerged with the rapid development of the Internet. These services are often of a single function. The value-added services can be achieved by combing with multiple services. The processing speed and stability of existing methods in service composition were not very well and seldom consider the fault diagnosis and handling methods for the service, which results in a greater probability of the service composition failure at run time. We use spiking neural P systems with colored spikes to model the fault of available service, component, and connector in the service composition. The proposed model can be used to locate a fault and handle it correctly when the service combination fails, the advantage of efficiency and stability of proposed method has been proved by comparing with the method of Petri net.