A multivariate empirical Bayesian (MEB) model with a dynamic intervention algorithm was developed to evaluate mean value, standard deviation and other quality parameters. This model takes full advantage of historical data which is smooth in characteristic and also information from other relevant points in error correction. This algorithm can remarkably reduce MEB error in the evaluation and simultaneously a series of error modes in manufacturing process are detected and formulated. Based on recognition of these error patterns, a dynamic intervention algorithm was developed. For validation and verification, this method was applied to the data from multipoints of a vehicle body and the result turns out to be satisfactory.