Native speakers of two languages (English and Ndonga) were compared on three colour cognition tasks (sorting, triads and visual search) in a test of the linguistic relativity hypothesis (Whorf, 1956). The colour lexicons of these two languages differ because Ndonga has no basic terms for ORANGE, PINK and PURPLE, and stimuli were chosen to exploit this difference. On the sorting task (sorting into similarity-groups) for each language, nominally similar colours were grouped together more often than nominally dissimilar colours. On the triads task (choosing the most different of three colours), when the most nominally isolated colour differed for the two language-groups, each group tended to choose their nominal isolate. On the search task (scanning for target colours among distractors), targets were either in a different English category than distractors (cross-category), or some distractors were in the same English category as distractors (within-category). The 'cost' in speed of having within-category distractors was much greater for the English than for the Ndonga. Overall, these data suggest that a core universal component is modulated by a small relativist influence. The differences in the visual search task are consistent with language affecting pre-attentive processes (an indirect language effect) as well as exerting on-line influences (a direct effect).