We report detailed observations of three pairs of expert scientists and a pair of advanced undergraduate students solving an experimental optics problem. Using a new method ("transition graphs") of visualizing sequences of logical steps, we were able to compare the groups and identify patterns that could not be found using previously existing methods. While the problem solving of undergraduates significantly differed from that of experts at the beginning of the process, it gradually became more similar to the expert problem solving. We mapped problem solving steps and their sequence to the elements of an approach to teaching and learning physics called Investigative Science Learning Environment (ISLE), and we speculate that the ISLE educational framework closely represents the actual work of physicists