The ability to learn, store, and retrieve information about relationships is impaired in schizophrenia. Here, we tested 38 control and 61 schizophrenia subjects for their ability to identify the novel pairing of stimuli, based on associations learned during training. Subjects were trained on 3 sets of paired associates: 30 face-house pairs (H-F1), 30 face-house pairs (H-F2, same house with new face), and 30 face-face pairs (F3-F4). After training, participants were tested on the 3 explicitly trained pair types, as well as 30 new face-face pairs (F1-F2), which could only be linked together via the same house during the H-F1/H-F2 training blocks. Of 99 subjects tested, 37 patients with schizophrenia and 36 age-matched healthy control subjects learned the premise pairs and performed the relational memory test. Healthy control subjects were significantly more accurate in identifying the inferential (F1-F2) pairs than the noninferential (F3-F4) pairs. In contrast, schizophrenia patients were equally accurate on inferential and noninferential pairs, providing evidence for a relational memory deficit in schizophrenia. However, the current version of the associative inference paradigm, suggested by the Cognitive Neuroscience Treatment Research to Improve Cognition in Schizophrenia initiative, has limited feasibility, calling into question the generalizability of the findings for the larger schizophrenia population.