Research on quantitative estimation often relies either on formal computational models tested with simplified laboratory stimuli or on informal theories applied to real-world tasks. The present study connects these approaches by testing established computational models with 80 real-world stimuli from three domains: food, countries, and mammals. A total of N = 142 participants were trained to estimate one of three domain-specific continuous criteria (carbohydrates per 100g for foods, life expectancy for countries, days to female maturity for mammals) and then judged new items in a test phase. We compared five candidate models (the Cue Abstraction Model, the Generalized Context Model, the RulEx-J hybrid, the Mapping Model, and the QuickEst heuristic) in their ability to account for participants' estimates in these domains using a simulation-based Bayesian model comparison. All models except QuickEst accounted for participants’ estimates, although the dominant cognitive processes differed across domains. These findings reveal domain-specific variation in estimation strategies and demonstrate that computational models originally developed for artificial tasks can also capture judgments in naturalistic settings. Overall, the results highlight the importance of testing cognitive process models beyond controlled laboratory contexts to better understand real-world quantitative estimation.