Aims: Stroke disrupts neural communications in the brain. The resulting network disruption may cause disabilities after stroke. The aim of this study was to examine whether the remaining networks can underlie a higher-order cognitive process in stroke. Methods: Fifty-eight stroke patients were studied using a measure of the intellectual capacity to reason by spatial analogy (Raven’s Coloured Progressive Matrices [RCPM]), and functional connectivity revealed by resting-state functional magnetic resonance imaging (rfMRI). A connectome-based analysis was performed on rfMRI data to examine whole-brain networks. In addition to empirical covariance matrices, group-sparse inverse covariance (GSIC) matrices were estimated using a method of group-sparse modelling and negated to get partial correlations, which infer direct connections. Correlations between the connectivity in the matrices and RCPM scores were examined, and a data-driven clustering analysis was performed. Results: A significant inverse correlation with the scores was found in the connectivity between the posteromedial and the left inferior parietal cortices in the negative GSIC matrices. In fact, as the connectivity increased, the scores decreased across three groups determined by clustering. The group with the highest connectivity showed a low-scoring pattern in a set whose items required apprehension of discrete figures as spatially related wholes. The lesions of the group were localized in the left subcortex. Conclusions: The remaining medio-lateral parietal connectivity may underlie reasoning by spatial analogy in stroke. The specific low-scoring pattern of distribution of the scores characterized the severely impaired spatial reasoning in the group with the highest parietal connectivity among groups.