Abstract: Freeway Incident Detection（AID） system as a very important part of Intelligent Transportation System，through discovering incidents that occur in freeway in time and trying the best to decrease the negative influence of incidents，can reduce traffic jam，guarantee freeway safety and reduce environment pollution.The paper introduces a strong and powerful classify tool—Support Vector Machines（SVM） to carry out freeway incident detection.Considering training data playing different roles in SVM and some isolate point or noise existence，we take a modified fuzzy-c-cluster method to pretreat the training data which can highly lessen training data and promote the training speed of SVM，and moreover，it has a favorable robustness.Simulation experiment shows the algorithm’s feasibility and validity.