The accuracy of saccade localisation during visual search was examined for a search target defined by the single features of orientation or depth or by a conjunction of the two features. Subjects were required to move their eyes to the target which appeared in one of eight possible locations, arranged circularly around fixation, with non-targets filling the remaining seven positions. Search for a target defined by a single feature resulted in approximately 70% correct first saccades in both cases, while search for the conjunction target resulted in only 40% correct first saccades. Furthermore, averaged latency for conjunction search was longer than for simple search. Nevertheless, some subjects showed a remarkably good ability to locate a conjunction target with a single saccade of short latency. An analysis of first saccades in terms of their speed and accuracy indicates that the target selection is not preceded by a covert scanning of the display but rather is a result of parallel processing of the visual information provided. We also relate our study to the study of conjunction search reported by Nakayama and Siverman [Nakayama, K., & Silverman, G.H. (1986). Serial and parallel processing of visual feature conjunctions. Nature, 320, 264-265.].