Whether activity in the autonomic nervous system differs during distinct emotions remains controversial. We obtained continuous multichannel recordings of autonomic nervous system activity in healthy adults during a video-based emotional reactivity task. Dimensionality reduction revealed five principal components in the autonomic time series data, and these modes of covariation differentiated periods of baseline from those of video-viewing. Unsupervised clustering of the principal component time series data uncovered separable autonomic states that distinguished among the five emotion-inducing trials. These autonomic states were also detected in baseline physiology but were intermittent and of smaller magnitude. Our results suggest the autonomic nervous system assembles dynamic activity patterns during emotions that are similar across people and are present even during undirected moments of rest.