For the complex product design knowledge with a large number catalogs and data volumes, the multi-classification problem was solved by using linear approach based on the margin and probability information of MPM. The true mean and covariance were replaced by the sample ones. The weight voting matrix was established, the coding matrix varied with the difference algorithm and training data. The classifier with max value was selected by using the amplitude value. Linear classifier was used as binary one for high speed. The angle difference measurement of subspace was proposed to characterize the difference of every class conception. The feasibility and practicality of the above mentioned method were approved by forklift product design knowledge.