Disclosed is an SVM-based multivariate quality diagnosis classifier for a manufacturing process. Original data of quality characteristics in the manufacturing process is collected and pre-processed, ahybrid algorithm is applied for conducting process analysis on multivariate quality characteristics of a key procedure, stability and anomalies are judged according to data recorded by a control chart, and a support vector machine method is applied for finding out where a source of the process anomalies is, so that a classification result is more accurate, and relaxation factors are added in an objective function by means of a Lagrangian optimization method. The classifier has the advantages that the process capability coefficient condition is strict, the judgement state is accurate, the algorithm complexity is low, the processing time is short, multivariate quality, misjudgment factors and principal component factors are integrated, the applicability is higher, parameter processing is standardized, data processing is improved, the misjudgment probability is reduced, the problems of data offset and unit inconsistency are solved, and an anomaly diagnosis technology can be achieved.