From molecules to whole organisms, the dynamics of natural living systems depart from “thermodynamic equilibrium” [Gnesotto, 2018]. Statistically, the state trajectories produced by such systems are non-reversible, i.e. they do not have equal likelihood of occurring in the reverse temporal direction even in stationary regimes. For example, neural systems exhibit anisotropic waves of activity (e.g. during development), produce precisely ordered spike sequences (e.g. episodic memory), or generate rotational patterns of activity (e.g. motor control). Despite the prevalence and importance of sequential neural activity, there is a relative paucity of methods for exploring the spatio-temporal structure of irreversibility in multivariate time series. Here, we introduce Sequential Components Analysis (SCA), a simple yet effective and scalable method for doing this. SCA performs a systematic analysis of spatio-temporal covariances (all time pairs and unit pairs), and extracts the spatio-temporal modes of activity that contribute most to sequentiality. We highlight the main distinguishing features of the method using a toy example, and apply it to monkey M1 motor activity as well as rat hippocampal data where we show that sequential features separate navigational memories better than mere principal components.