The paper introduces a new method for analyzing functional magnetic resonance imaging (fMRI) data using independent component analysis (ICA). The authors decompose fMRI data from four normal subjects performing various tasks into spatially independent components, each with a unique time course of activation. ICA extracts spatially independent components from 144 time points collected during a 6-minute trial, and in all trials, one component closely matches the 40-second alternations between experimental and control tasks. These task-related components overlap with those detected by standard correlational analysis but include additional frontal regions. Other ICA components show transient, quasiperiodic, or slowly varying activation. Higher-order statistics enforce stricter criteria for spatial independence, making ICA and a related fourth-order decomposition technique superior to principal component analysis (PCA) in determining the spatial and temporal extent of task-related activation. The ICA method is robust to simulated noise and can distinguish between task-related, nontask-related, and artifact signals. The technique is demonstrated through a "thought experiment" comparing ICA, PCA, and correlation analyses, and applied to fMRI data from subjects performing the Stroop color-naming and word/number tasks. The results show that ICA accurately identifies consistently task-related components and provides more precise spatial and temporal information compared to PCA and correlation methods.The paper introduces a new method for analyzing functional magnetic resonance imaging (fMRI) data using independent component analysis (ICA). The authors decompose fMRI data from four normal subjects performing various tasks into spatially independent components, each with a unique time course of activation. ICA extracts spatially independent components from 144 time points collected during a 6-minute trial, and in all trials, one component closely matches the 40-second alternations between experimental and control tasks. These task-related components overlap with those detected by standard correlational analysis but include additional frontal regions. Other ICA components show transient, quasiperiodic, or slowly varying activation. Higher-order statistics enforce stricter criteria for spatial independence, making ICA and a related fourth-order decomposition technique superior to principal component analysis (PCA) in determining the spatial and temporal extent of task-related activation. The ICA method is robust to simulated noise and can distinguish between task-related, nontask-related, and artifact signals. The technique is demonstrated through a "thought experiment" comparing ICA, PCA, and correlation analyses, and applied to fMRI data from subjects performing the Stroop color-naming and word/number tasks. The results show that ICA accurately identifies consistently task-related components and provides more precise spatial and temporal information compared to PCA and correlation methods.