For multiblock batch process, the relationship of variables between two adjacent blocks was not fully considered by the conventional fault diagnosis method. Therefore, in this paper, an intermediate block dependency analysis (IBDA) reconstruction algorithm is proposed, based on which normal information is separated, and online fault detection along the fault direction is achieved. The main contributions are as follows: (1) IBDA is explored to analyze the relative change of variables between two adjacent blocks, and then the increased subspace, normal subspace, and decreased subspace in each block space are separated; (2) to improve the accuracy of fault correction, the fault directions are accurately selected by calculating the fault amplitude of each direction, as a consequence of which both the responsible variables and the uninformative variables are selected; (3) fault is reconstructed by calculating fault information along fault directions, and a reconstruction-based detection model is given, which is more effective than the conventional method. The advantage and effectiveness of the proposed method are illustrated with penicillin fermentation process.