For the multi-rate sampling problem of the penicillin fermentation process, some traditional methods cannot be solved due to the different lengths of samples. To solve this problem, a grouping multi-rate sampling fault detection method is proposed. The training sample set is divided into multiple groups according to different sampling rates of variables. Each group has the same sample length. We use the principle component analysis and k-nearest neighbour methods to build model for every group. Then the models of all groups are combined to determine the control limit. The PenSim2.0 software is used to generate a sample set of the penicillin fermentation process that contained multiple variables at different sampling rates. The simulation result shows the effectiveness of the proposed method.