A number of algorithms have been developed which predict the performance of human beings during visual search in complex, realistic scenes. The performance of these algorithms is generally poor in scenes that differ from the scenes for which they were developed and tested. Moreover, the algorithms do not give any insight in the mechanisms underlying visual search. In this study we measure the performance and eye movements of subjects who perform visual search tasks in complex scenes. We investigate the relationship between eye movement parameters such as fixation duration and saccadic amplitude and scene parameters.