Objective: To identify cognitive subgroups (comprising neurocognition and social cognition domains) within first-episode psychosis (FEP) patients including a healthy control group for comparison. Predictive validity of cognitive clusters in relation to symptoms and functioning was also investigated. Method: A comprehensive cognitive battery was administered to 133 FEP participants and 46 healthy controls. Ward’s method hierarchical agglomerative cluster analysis with k-means verification was used to determine clusters. Clusters were externally validated and 6-month predictive validity was also examined. Results: Three distinct clusters were identified and were defined by degree of impairment rather than specific deficit profiles. Social–cognitive performance mirrored neurocognitive performance in each cluster. Cluster 1 was characterized by significant widespread cognitive impairments (1–2 SD below the mean) and solely comprised FEP participants (n = 24). Cluster 2 suggested moderately impaired cognitive functioning (within 0.5 SD below the mean), and comprised mostly FEP participants and 2 healthy controls (n = 73). Cluster 3 showed a pattern of cognitively intact performance across domains and comprised 37 FEP participants and 44 healthy controls (n = 81). Premorbid IQ, negative symptom severity, and functioning were significantly associated with cluster membership at baseline. At 6-month follow-up, cluster membership remained significantly associated with negative symptoms and functioning. Conclusions: The heterogeneity of cognition in FEP may be based on degree of impairment across both neurocognitive and social–cognitive domains. Cognitive clusters were associated with symptom and functional outcome, suggesting that measurement of cognition at entry to treatment may be useful for prognosis and treatment.