The large-scale and complexity of modern chemical plants makes it difficult for the operator to constantly monitor all process variables. Numerous methods exist for monitoring processes; however most of them suffer from computational complexity and scale-up problems when applied to large-scale processes. Further, their accuracy also degrades when inessential variables are included in the analysis. In this paper, we describe a systematic method for identifying key variables for process monitoring. The proposed method is especially suited for multi-state processes that operate in different regimes at different times; in such cases different sets of key variables would be needed in different states. A classification of key variables into state-indication, state-differentiation, state-progression, active, external-effect and important-balance variables is proposed. This provides a systematic basis for defining state-specific key variables that reflect the unique and essential features of a state. Methods for identifying the different key variables using process flowsheet, operating procedure, and historical operations data have also been developed. The proposed methodology is illustrated on the startup transition of a simulated fluidized catalytic cracking unit. The benefits of key variable based process monitoring are reduction in monitoring load and improvement in sensitivity.