SVM has already shown its successful application on the CCP discovery for HACCP implementation. However, the classic non-incremental SVM method is not an effective algorithm due to the thoroughly re-study for those samples are gradually added. We propose a new incremental algorithm. It makes use of the heuristic mode so that training can be firstly applied on the cases with greater possibilities to be SVs, and the training set can be reduced. The experiments show that the training speed is visibly improved without losing the precision of the classification.