People accurately compare and estimate means without using formal calculations, however, little is known aboutthe cognitive processes underlying these behaviors. We used objective, behavioral data (e.g., eye fixation patterns), which arecompatible with multiple representations, to compare cognitive models. Specifically, we compared seven cognitive modelsincluding working memory activation (weighting values as a function of the number of and duration of fixations), workingmemory constraint (e.g., recency + primacy, last four), or Bayesian models (e.g., first fixation set as prior).Our task presented sets of 5 to 10 3-digit numbers (framed as the result of a home run derby) and asked participants to predicthow far the next ball would be hit. The same fixation data were loaded into each model to create a unique estimate, which wasthen compared to the participant’s actual prediction. The difference between the model and actual was calculated to create anaccuracy index.