With good eggs and crack eggs as experimental subjects,the machine vision and support vector machines （SVM) were used to study the differences between good eggs and crack eggs, and multi-feature parameters were extracted to achieve automatic recognition of crack eggs. Firstly, an algorithm would be run to eliminate bright spots on the preprocessed image of the surface of eggs before marking them by region. Secondly, 13 characteristic parameters from five different domains to identify the good eggs and crack eggs were extracted, and these parameters were as follows: the marked region parameters of images (the number of markers and the marker area points), the geometric parameters (the major axis and the minor axis), the shape parameters based on Freeman chain code (the shape number), the texture parameters (the mean, the standard deviation, the smoothness, the third moment, the uniformity and the entropy) and the spectral parameters (the maximum amplitude and the maximum phase). Thirdly, to highlight the greater impact factors between 13 parameters and to shorten the detection time, adaboosting algorithm was used to optimize the above parameters, which was the input vector of SVM. Finally, the recognition model was built by SVM. The results indicated that the accuracy rate of the recognition model was 97.5%, which could meet the requirements of enterprises basically.