In our daily lives we often make quantitative judgments based on multiple pieces of information such as evaluating a students paper based on form and content. Psychological research suggests that humans rely on several strategies to make multiple-cue judgments. The strategy that is used depends on the structure of the task. In contrast, recent research on learning in judgment tasks suggests that learning is relatively independent of task structure. In a simulation study we investigated how the performance of several learning models is influenced by the structure of the task and the amount of learning experience. We found that a linear additive neuronal network model performed well regardless of task structure and amount of learning. However, with little learning a heuristic model performed similarly well, and with extensive learning, associative learning models caught up with the linear additive model.