Multi-state operations are common in chemical plants and result in high-dimensional multivariate, temporal data. In this paper, we develop self-organizing map (SOM) based approaches for visualizing and analyzing such data. The SOM is used to reduce the dimensionality of the data and visualize multi-state operations in a three-dimensional map. During training, neuronal clusters that correspond to a given process state - steady state or transient - are identified and annotated using historical data. Clustering is then applied on SOM to group neurons of high similarity into different clusters. Online measurements are then projected on to this annotated map so that plant personnel can easily identify the process state in real-time. Modes and transitions of multi-state operations are depicted differently, with process modes visualized as a cluster and transitions as trajectories across SOM. We illustrate the proposed approach using data from an industrial hydro-cracker. © 2008 Elsevier B.V. All rights reserved.