This article presents insights from a laboratory experiment on human problem solving in a combinatorial task. I rely on a hierarchical rugged landscape to explore how human problem-solvers are able to detect and exploit patterns in their search for an optimal solution. Empirical findings suggest that solvers do not engage only in local and random distant search, but as they accumulate information about the problem structure, solvers make ‘model-based’ moves, a type of cognitive search. I then calibrate an agent-based model of search to analyse and interpret the findings from the experimental setup and discuss implications for organizational search. Simulation results show that, for non-trivial problems, performance can be increased by a low level of persistence, that is, an increased likelihood to quickly abandon unsuccessful paths.