This work combines model-based and process-tracing analyses to investigate exemplar retrieval in numeric judgments from multiple cues. Specifically, we test whether the reliance on exemplars indicated by cognitive modeling is reflected in the gaze proportions to the exemplar locations on the screen. We conducted two eye tracking studies (Ns = 19 and 49), in which participants judged the criterion value of objects with two multi-valued features, related to the criterion in an additive (Study 1) or multiplicative (Study 2) way. Participants first learned the criterion values and locations of four exemplars, each presented in a different corner of the screen, and then judged the criterion value of new, briefly presented transfer stimuli, varying in their similarity to the exemplars. To test the exemplar retrieval during transfer, eye tracking measured the gaze proportion to the now empty screen corners (looking-at-nothing behavior), and cognitive modeling using the RulEx-J framework modeled the reliance on exemplars. Consistent with previous research, cognitive modeling and eye tracking suggest more exemplar retrieval in the multiplicative than in the additive task. Focussing on the multiplicative task, we found that participants relying more on exemplars also look more to the exemplar locations (r = 0.36, p = .01). Within-trial analyses further showed that these participants look in particular at exemplars similar to the transfer stimulus. By making use of the close link between cognitive modeling and eye tracking, our results provide new insights into the nature of exemplar retrieval in inferences from multiple cues.