The invention discloses a suspect early warning method based on face recognition. The suspect face image eigenvalue is calculated by extracting the suspect face image. The eigenvalues of suspects' face images are compared with those in the blacklist database filed by the public security bureau. Judging whether the characteristic value of the suspect's face image is recorded in the blacklist database filed by the public security bureau, if so, sending an alarm signal to the driver, uploading the alarm signal to the dispatching monitoring center, and sending the alarm information to the public security bureau; if not, the eigenvalue comparison analysis is performed again. The invention also discloses a suspect early warning system based on face recognition, comprising a face image extractionmodule, an eigenvalue comparison module, a suspect judgment module and a monitoring alarm module. The invention realizes high-precision and high-efficiency face recognition, facilitates the identification of suspect basic information, understands suspect personnel activity, and can real-time call monitoring video, on-line video-on-demand, saving bandwidth resources.