Individuals with bipolar disorder (BD) show heterogeneity in clinical, cognitive, and daily functioning characteristics, which challenges accurate diagnostics and optimal treatment. A key goal is to identify brain-based biomarkers that inform patient stratification and serve as treatment targets. The objective of the present study was to apply a data-driven, multivariate approach to quantify the relationship between multimodal imaging features and behavioral phenotypes in BD. We pooled structural, task and resting-state functional magnetic resonance imaging (MRI), and clinical, cognitive, and functioning data from 167 fully or partly remitted patients with BD from three studies conducted at the same site. We performed canonical correlation analysis (CCA) to investigate multivariate relations among the 56 imaging and 23 behavioral features in patients. Data from 46 matched healthy controls were included for covariate-adjusted standardization of patients' scores and for group comparisons. The imaging and behavioral data sets showed a strong canonical correlation (r = 0.84, p = .004). Among the behavioral variables, cognitive test scores across psychomotor speed, verbal memory, and verbal fluency were associated with the multimodal imaging variate comprising task activation within the dorsolateral prefrontal cortex and supramarginal gyrus, also when other clinical and daily functioning variables were considered. Task activation within the dorsal prefrontal and parietal cognitive control areas constitutes a potential pro-cognitive treatment target.