Abstract Amyotrophic lateral sclerosis (ALS) is a progressive neurodegenerative disease characterized by the loss of motor neurons in primary motor cortex, leading to muscle weakness, atrophy, and death within a median of three years. Even though ALS is characterized by different disease subtypes affecting different body parts, individualized phenotyping of functional ALS pathology has so far not been achieved. We recorded 7 Tesla functional MRI data while ALS patients and matched controls moved affected and non-affected body parts in the MR scanner. We applied robust Shared Response Modeling for capturing ALS-specific shared responses for group classification, and Partial Least Squares regression for relating the latent variables to clinical subtypes and the degree of disease progression. We show that disease onset and severity can be best modeled by functional connectivity rather than local activation changes. We also show that functional disease-defining information in primary motor cortex is not strongest in the area that is behaviorally first-affected, deviating from the behavioral phenotype of the patients. When computing the model’s weight distribution of the King stage classification and projecting them back into voxel space, the highest mean weights are present in the foot and tongue/face regions. Our data highlights the importance of 7 Tesla functional MRI task-based functional connectivity measures for classifying ALS-patients in addition to structural readouts,and provides evidence that a 7 Tesla functional MRI can be used for identifying a disease signature of each individual ALS patient.