Sensory perception relies on the brain's integration of multiple imprecise inputs, a process known as cue combination. Previous research has investigated multisensory integration or unisensory integration within non-haptic senses. In contrast, cue combination within the sense of touch has been understudied. Here, we investigated whether humans optimally combine haptic cutaneous and finger configuration cues when discerning the size of a disk held edge-on between the thumb and index fingers. When these two fingers span the disk to contact its perimeter, a finger configuration cue (the perceived distance between the fingers) provides information about the disk's size. Less obviously, cutaneous cues to disk size may result from the indentation of finger pads caused by the disk's curvature. We considered three hypotheses for how humans might use these cues: they might rely solely on the most reliable cue (Winner-Take-All Model, WTA), combine cues based on a simple arithmetic average (Average Model, AVG), or combine cues via an optimal weighted average (Optimal Model, OPT) in which more reliable cues exert proportionately greater influence on the percept. In three experiments involving 34 participants, we compared participant performance to the predictions of the three models. Each experiment tested participants using a two-interval forced-choice (2IFC) paradigm with 3D printed disk stimuli. On each trial, under occluded vision, participants felt two disks sequentially and responded which felt larger. Participants were tested with the cutaneous index finger cue, cutaneous thumb cue, and finger configuration cue individually and with the three cues together. The three experiments were designed to have progressively greater resolution to distinguish the relevant models from one another. In Experiments 1 and 2, the disks were circular. Experiment 3 included non-circular cue-conflict stimuli. The improvement of accuracy in multi-cue conditions, and perceptual effects of cue-conflict stimuli, were broadly consistent with optimal cue combination. Eight of 12 participants were classified as OPT in Experiment 1; eight of 11 in Experiment 2; and 10 of 11 in Experiment 3. The mean confidence of OPT classifications was 0.56, 0.61, and 0.98, respectively. We conclude that humans combine haptic cues optimally to judge the sizes of held objects.