Abstract Chemical processes usually work under different operating modes to meet the market demand and to achieve higher profits. This necessitates investigating related algorithms and alarm systems for multimodal chemical processes. Although some research effort has been made to monitor multimodal processes, little attention was paid to fault diagnosis issue when addressing multiple modes. In this paper, we present both a label consistent dictionary learning (LCDL) based multimode process monitoring approach and sparse contribution plot (SpCP) for fault diagnosis. Firstly, a discriminative and reconstructive dictionary is obtained from normal historical process data via label consistent K-SVD algorithm. In addition, we augment the learned dictionary to get another dictionary, which consists of two blocks, one for multiple normal operating modes and another for faults. Then, during online monitoring period, a new sample is coded sparsely using the aforementioned augmented dictionary. After that, its dictionary reconstruction residual (DRR) is calculated for fault detection purpose. At last, a novel sparse contribution plot is proposed to figure out the root cause of the detected fault. The SpCP is better able to highlight the real cause with no ambiguity in that only a small fraction of variables’ sparse contributions are nonzeros. The effectiveness of the proposed methodology is demonstrated by both a numerical simulation and a continuous stirred tank heater (CSTH) process.