The invention discloses a machine vision detection method based on four-colour printed matter registration detection marks, comprising the following steps of: presetting graphs of the four-colour printed matter registration detection marks, and taking the circle centre of a black solid circular mark graph as a reference of the registration detection marks, wherein the circle centre of a yellow annular mark graph, the circle centre of a pinkish red annular mark graph, and the circle centre of a cyan annular mark graph are respectively as same as the circle centre of the black solid circular mark graph; acquiring images of preset graphs of the registration detection marks during a printing process; respectively obtaining coordinates of the circle centres of the registration detection marks in the images by a computer system; calculating registration deviation via the coordinates of the circle centres; comparing the squared value of the registration deviation with a deviation tolerance level; and finally, outputting a registration deviation result. The registration detection marks are simple in structure and small in occupied space, and camera can be convenient to rapidly and accurately shoot. According to the corresponding detection method, only deviation analysis does need to be performed via circle centre parameters, thus being high in detection efficiency, intuitive in result, and capable of realizing online detection for high-speed printed matters.