The block design task, a standardized test of nonverbal reason-ing, is often used to characterize atypical patterns of cognitionin individuals with developmental or neurological conditions.Many studies suggest that, in addition to looking at quantita-tive differences in block design speed or accuracy, observingqualitative differences in individuals’ problem-solving strate-gies can provide valuable information about a person’s cogni-tion. However, it can be difficult to tie theories at the levelof problem-solving strategy to predictions at the level of ex-ternally observable behaviors such as gaze shifts and patternsof errors. We present a computational architecture that is usedto compare different models of problem-solving on the blockdesign task and to generate detailed behavioral predictions foreach different strategy. We describe the results of three differ-ent modeling experiments and discuss how these results pro-vide greater insight into the analysis of gaze behavior and errorpatterns on the block design task.